You can use the Gemini API, GroqCloud and OpenRouter’s free models without paying or adding a card, but almost no AI company gives individuals a free dollar credit for its API at sign-up (checked 12 and 13 October 2026). What you get instead is a free tier: an ongoing allowance with no end date, capped by a rate limit. A rate limit is a ceiling on requests, or on tokens (the chunks of text a model reads and writes), per minute or per day. Some free tiers are paid for with your data, which matters if you work with client files.
Startup API credits are a different thing. Credits are a prepaid balance the provider draws down as you use it; they are not cash, and most expire. OpenAI’s startup credits need a referral code from a partner VC (a venture capital firm in OpenAI’s partner network), and Claude for Startups is not taking new applications (its page said on 15 October 2026 that the current cohort is full). This page covers the 40 free tiers, startup programs and GPU credits we track for AI APIs, each checked on its official page, with every limit dated.
Looking for free ChatGPT or Gemini app plans rather than the API? See free ChatGPT, Gemini and Claude plans.
Gemini API
No card
Free tier in AI Studio. Exact limits shown only inside AI Studio.
GroqCloud Free
1,000 a day
Requests on gpt-oss-120b, 30 a minute.
Kaggle
30 GPU hours
A week, plus 20 TPU hours.
Free AI APIs at a glance (October 2026)
These are the AI model APIs, often called LLM (large language model) APIs, that had a free allowance or trial on 12 and 13 October 2026: what the free part covers and what you give in return. Two rows are here so you can rule them out. Hugging Face gives free accounts no inference credit (credit for running models), and OpenAI’s daily tokens need money on the account first. Each provider name links to its full entry with the official limits page. We link to those pages rather than copying full rate tables, because limits change often.
| Provider | Free allowance | Card | Your data | Production use |
|---|---|---|---|---|
| Gemini API | Free tier on Gemini 3.x Flash and Flash-Lite, 2.5 Pro and Flash, Gemma 4 and embeddings. Exact limits shown only inside AI Studio. Daily quotas reset at midnight Pacific time | No | Used to improve Google products; may be read by human reviewers | Not stated. In the EEA, Switzerland and the UK, only paid services may be used in apps for users there |
| GroqCloud Free | gpt-oss-120b: 30 requests a minute, 1,000 a day, 8K tokens a minute, 200K tokens a day | No | Not stated | Not stated. Higher limits, Batch and Flex need the paid Developer plan |
| OpenRouter :free models | 20 requests a minute and 50 a day; 1,000 a day once you have bought at least 10 credits | No | Providers that log prompts are used only if you turn on the model-training toggle | Not stated. The free model list changes often |
| Mistral Free mode (medium confidence) | The lowest rate limits; exact figures only on the Limits page in the Admin panel | No | An “Anonymous improvement data” toggle under Privacy in the Admin panel | Meant for evaluation and prototyping |
| Cohere Trial key | Free, rate-limited calls; limits not shown on the pricing page | Not stated | Not stated | No: not for production or commercial use |
| NVIDIA API Catalog (medium confidence) | Free endpoints on 38 of 97 catalogue models. The site’s embedded page copy, not a published FAQ, gives up to 40 requests a minute | Not stated | Not stated | No: development and testing. Production goes to NVIDIA AI Enterprise or paid partner endpoints |
| Cloudflare Workers AI | 10,000 Neurons a day (Cloudflare’s unit of Workers AI usage), reset at 00:00 UTC | No | Not stated | Not stated. On the Free plan, Workers AI stops at the daily limit |
| Cerebras trial | $5 once, for 30 days: 5 requests a minute and 1M tokens a day on gpt-oss-120b and qwen-3.8-27b. No renewing free tier | Yes, a payment method | Not stated | Not stated. Access pauses when the credit expires or runs out |
| Hugging Face | No monthly inference credit on free accounts. PRO costs $9 a month and includes $2 | Free account: no. PRO is a paid plan | Not stated | Not stated |
| OpenAI data-sharing tokens | Build tier: 250K flagship and 2.5M mini tokens a day. Launch and Grow: 1M and 10M. Reset at 00:00 UTC | Yes: you need a positive account balance | You share inputs and outputs, which OpenAI can use to train models | Not stated. OpenAI decides which organisations are eligible |
How to read it. Gemini, GroqCloud, OpenRouter, Mistral and Cloudflare are ongoing free tiers: no end date, low limits. Cerebras is a one-off trial, and it needs a payment method. Cohere’s and NVIDIA’s free calls are not for production use.
Small print that costs money. OpenRouter charges a 5.5% fee ($0.80 minimum) when you buy credits by card, and its terms let it expire unused credits a year after purchase. Workers AI stops at the daily limit on the Workers Free plan; using more means Workers Paid, at $5 a month minimum. Mistral’s limits rise only with billed usage, not with prepaid credit you buy.
From India. India is on the Gemini API’s available-regions list and on OpenAI’s API supported-countries list. Groq’s billing FAQ has special billing thresholds for Indian billing addresses. Cerebras needs a card, so an Indian card must have international payments turned on.
Free AI APIs and credits to try first
Six options we would start with. We confirmed the key details of all six on official pages on 12 and 13 October 2026.
Most capable free tier, no cardGemini API free tierGoogleFree, rate-limited Gemini Flash, 2.5 Pro and GemmaCovers Gemini 3.x Flash and Flash-Lite and 2.5 Pro. Free-tier prompts may be read by human reviewers, so use test data.Fast open models, no cardGroqCloud Free planFree tier: 1,000 requests a day on gpt-oss-120bNo end date. The cap is 30 requests a minute and 200K tokens a day on that model.Many models, one keyOpenRouter free modelsFree models: 20 requests a minute, 50 a dayOne small purchase raises the daily cap twentyfold. The free model list changes often.Already paying OpenAIComplimentary daily tokens for sharing API data with OpenAI250K flagship + 2.5M mini tokens a day (Build tier)Only if OpenAI shows your organisation the offer. Shared inputs and outputs can be used to train its models.Free GPU notebooksKaggle Notebooks free GPU and TPU quota30 GPU hours + 20 TPU hours a weekOne P100 or two T4s, up to 12 hours a session. Expect a queue at busy times.GPU credits for startupsTogether AI Startup AcceleratorUp to $15,000 credits (Build tier, up to $5M raised)For startups that raised up to $5M. Selection-based, and the credits do not cover Reserved GPU Clusters.
Fake “free API key” sites: how to spot them
Search for a free OpenAI or Claude API key and you will find sites, videos and chat groups offering one. None of them is an official route. These are the patterns to walk away from:
- Shared or leaked keys. Lists of keys “that still work” belong to someone else’s account. The owner pays, and the provider can revoke them at any moment.
- Key sellers. A “lifetime” or “unlimited” key sold for a small fee is either stolen or a resold proxy.
- Telegram or WhatsApp bots that hand out keys or “free premium model access” after you share a link or join a channel.
- Proxies that read everything you send. A free endpoint that “works just like the OpenAI API” passes your prompts and files through a server you know nothing about.
- Sites that ask for your own key or card to “activate” free credit, “verify” your account or “double” your balance.
The simple rule: a key you did not create in the provider’s own console is someone else’s key or a proxy. Never paste your key into a site promising credit. Create keys only in official consoles: Google AI Studio, the GroqCloud console, OpenRouter, Mistral Studio, the OpenAI platform and the Claude Console.
If you have already pasted a key somewhere, delete it in the provider’s console and create a new one. Then check your usage page for calls you did not make.
Free AI API tiers, notebooks and coding agents
Every free AI API tier, GPU notebook and AI coding agent we track, plus the API credits that come with paid plans, A to Z. Each entry gives the exact allowance, the card rule, an India note and the official link, checked 12 and 13 October 2026. Tiers that have ended are hidden until you choose Show ended and paused. They then appear in a muted list at the end, so you can recognise them in old tutorials. Use No card needed to see the ones you can start without a card.
Show only
Status
No programs match these filters.
Skip past this listCerebras Inference Free Trial
CerebrasFree tier replaced by a $5 trialAdd a payment method to activate
Solo developersAnyone
$5 one-time credit for 30 days
One-time $5 credit when you add a payment method, valid 30 days. Trial limits: 5 requests a minute and 1M tokens a day on gpt-oss-120b and qwen-3.8-27b.
Eligibility, card, India and how to apply for Cerebras Inference Free Trial
- Who qualifies
- Add a valid payment method.
- One trial per person; credits can be revoked for fraud, abuse or duplicate accounts.
- How long
- 30 days after activation, or until the $5 is used.
- Referral or funding needed
- No
- Card needed
- Yes
- Work or school email
- No
- Countries
- No country limit stated on the pages checked.
- India
- Yes. No country restriction stated; an Indian card needs international payments enabled.
- How to apply
- Sign up at cloud.cerebras.ai and add a payment method to receive the credit.
- Watch out for
- API and Playground access pause when the credit expires or runs out, unless you buy PayGo credits.
- You are not charged or moved to paid usage automatically.
- Status note
- On 12 October 2026 Cerebras’s docs said there is no free tier that renews or any always-free allowance; the Free Trial is limited by time and credit. The pages do not say when this replaced the earlier free tier. Your first credit purchase moves you to the Developer tier.
Official page for Cerebras Inference Free Trial (opens in a new tab)Checked 12 October 2026
Cohere Trial API key
CohereActiveNo application: free trial key
Solo developersStudentsResearchers
Free, rate-limited Trial API key
Every new Cohere account gets a Trial API key with free, rate-limited API calls for testing and learning.
Eligibility, card, India and how to apply for Cohere Trial API key
- Who qualifies
- Cohere account (a Trial key is created automatically)
- Not for production or commercial use
- How long
- Ongoing
- Referral or funding needed
- No
- Card needed
- Not stated
- Work or school email
- No
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated on the pricing page; we did not confirm India sign-up on an official page.
- How to apply
- Create a Cohere account; the Trial key appears on the dashboard under API Keys.
- Watch out for
- Trial keys may not be used for production or commercial purposes.
- Production keys are billed pay-as-you-go, invoiced at month end or when the balance reaches $250.
- Status note
- No public Cohere startup credit program found: cohere.com/startup-program redirects to the home page (12 October 2026). Trial rate-limit figures are not shown on the pricing page.
Official page for Cohere Trial API key (opens in a new tab)Checked 12 October 2026
Complimentary daily tokens for sharing API data with OpenAI
OpenAIActiveEligible orgs opting in to data sharing
Solo developersBootstrapped startupsFunded startups
250K flagship + 2.5M mini tokens a day (Build tier)Best case: 1M + 10M tokens a day on Launch/Grow tiers
Eligible API organisations that share inputs and outputs with OpenAI get free tokens every day. Launch/Grow tiers get 1M on flagship models and 10M on mini models; Build tier gets 250K and 2.5M. Resets 00:00 UTC.
Eligibility, card, India and how to apply for Complimentary daily tokens for sharing API data with OpenAI
- Who qualifies
- Organisation owner must opt in to share inputs/outputs (all projects or selected ones)
- The offer appears on the data-sharing settings page only for eligible organisations
- Zero Data Retention organisations and Enterprise customers cannot opt in to input/output sharing
- A positive account balance is required to use the models and the free tokens
- How long
- Daily allowance, ongoing; OpenAI will give 30 days’ notice before ending it
- Referral or funding needed
- No, but OpenAI decides which organisations are eligible
- Card needed
- Yes
- Work or school email
- No
- Countries
- Countries where OpenAI’s API is supported (India included); no other country rule stated. Eligibility is shown per organisation.
- India
- Yes. India is on OpenAI’s API supported-countries list (checked 13 October 2026), and the help article sets no country rule. OpenAI still decides eligibility for each organisation, so check your data-sharing settings page for the eligibility banner.
- How to apply
- As an org owner, open the API platform’s data sharing settings. If you see the eligibility message, enable input/output sharing and confirm the ‘You’re enrolled for complimentary daily tokens’ message.
- Watch out for
- Shared data can be used to evaluate and train OpenAI models. Do not send sensitive, confidential or proprietary data.
- If one request would push the day’s total over the quota, the whole request is billed.
- Fine-tuned models, fine-tuning training, evals and tool use are excluded.
- Only traffic on projects with sharing enabled counts.
- The quota is shared across all models in each group.
- Status note
- Help article showed ‘Updated: 4 days ago’ on 12 October 2026 and lists GPT-6.x, GPT-5.x and older models in the two groups. Usage tiers in the article are named Build, Launch and Grow. Since 6 March 2025 tokens refresh at 00:00 UTC. Evaluations shared with OpenAI are processed at no cost for up to 7 runs a week; shared reinforcement fine-tuning jobs get discounted inference rates. (Corrected 13 October 2026 after a second check.)
Gemini API free tier (Google AI Studio)
GoogleActiveNo application; 18+, supported region
Solo developersStudentsBootstrapped startupsAnyone
Free, rate-limited Gemini Flash, 2.5 Pro and Gemma
Free tier for the Gemini API and AI Studio on many current models, including Gemini 3.x Flash and Flash-Lite, Gemini 2.5 Pro and Flash, Gemma 4 and embeddings, with lower rate limits. Pro previews and image or video models are paid-only. Gemini 4 Argon is paid-only: Google says it is not available on the Free tier or on the Google AI Pro plan (pricing page updated 14 October 2026).
Eligibility, card, India and how to apply for Gemini API free tier (Google AI Studio)
- Who qualifies
- Age 18 or over.
- An active Google Cloud project (or free trial); no billing account needed for the Free tier. Tier 1 needs a linked billing account.
- Must be in an available region.
- How long
- Ongoing; daily quotas reset at midnight Pacific time.
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- Available regions only. India, the UK and all EU countries are listed; China, Russia and Hong Kong are not.
- India
- Yes. India is on the official available-regions list.
- How to apply
- Sign in to aistudio.google.com and create an API key.
- Watch out for
- Free-tier prompts and outputs are used to improve Google products and may be read by human reviewers. Do not send sensitive data.
- In the EEA, Switzerland and the UK, the paid-service data terms apply even to free use, and only paid services may be used in apps you make available to users there.
- Google Search grounding on Gemini 3 and newer models is not available on the free tier. The 5,000 free grounding requests a month apply to paid use.
- Paid prices for Gemini 3.8 Flash and 3.6 Flash (and their TTS models) double from 1 January 2027.
- The Google Cloud $300 trial credit cannot pay for Gemini API use in AI Studio.
- Status note
- The free model list changes often; the pricing page was last updated on 14 October 2026 (checked 15 October 2026) and lists Gemini 4 Argon as paid-only. Gemini 2.5 Flash Image (paid-only) was scheduled to shut down on 2 October 2026. Exact free rate limits are shown only inside AI Studio. API terms last modified 28 April 2026. [Also: Prepay and Postpay billing plans took effect on 23 March 2026; new paid users default to Prepay (minimum $5, maximum $5,000 top-up). Batch and Flex are generally not on the free tier.]
Official page for Gemini API free tier (Google AI Studio) (opens in a new tab)Checked 15 October 2026
GitHub Copilot Free
GitHubChanged 1 June 2026No application; individuals only
Solo developersStudentsOpen-source maintainersAnyone
2,000 completions a month + small chat allowance
Free Copilot for individuals: 2,000 code completions a month plus a small monthly allowance of GitHub AI Credits for chat and agent use.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for GitHub Copilot Free
- Who qualifies
- Individual developers without Copilot through an organisation or enterprise.
- Holders of a Copilot Business or Enterprise seat are not eligible.
- How long
- Ongoing; the allowance resets at 00:00 UTC on the 1st of each month.
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated.
- India
- Yes. No country restriction stated.
- How to apply
- Turn on Copilot Free from your GitHub account settings or in a supported editor.
- Watch out for
- Unused credits do not roll over.
- No cloud agent on Free, and code review is limited to ‘Review selection’ in VS Code.
- Copilot Pro costs $10 a month and includes 1,500 AI credits (1,000 base + 500 flex).
- Copilot Pro free trials were paused in April 2026; we found no notice that they have restarted.
- Status note
- Since 1 June 2026, all Copilot plans bill in GitHub AI Credits (1 credit = $0.01). GitHub’s docs do not state the Free allowance; the plans page FAQ still says 2,000 completions and 50 chat requests a month, while its table says ‘limited chat and agent usage’. Docs say Free gets auto model selection only, while the plans page lists a broad model choice. GitHub paused new Copilot Pro trials from 10 April 2026 and paused new Student, Pro and Pro+ sign-ups for a time (still paused on 1 June); paid sign-ups began reopening gradually from 17 June 2026. GitHub said Copilot Free stayed available throughout. (Corrected 13 October 2026 after a second check.)
Official page for GitHub Copilot Free (opens in a new tab)Checked 13 October 2026
Google Antigravity Individual plan (free), including Antigravity CLI
GoogleActiveNo application; 18+, supported country
Solo developersStudentsAnyone
Free agent IDE + CLI with a weekly Gemini quota
A $0 Individual plan for Google’s agent-based IDE and the Antigravity CLI: unlimited Tab completions and Command requests, plus Gemini agent models up to a weekly limit. It replaced the free Gemini CLI and Code Assist tier.
Eligibility, card, India and how to apply for Google Antigravity Individual plan (free), including Antigravity CLI
- Who qualifies
- Aged 18 or over (the FAQ says Antigravity is unavailable to under-18s; adults can verify age with Google)
- Personal Google account (the FAQ suggests a @gmail.com address if a Workspace account gives problems)
- Country must be on Antigravity’s supported list (check the country on your Google Terms of Service page)
- How long
- Ongoing; agent-model quota refreshes weekly
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- Available only in the listed countries and territories: Americas 51, Europe 47, Africa 56, Asia 43, Oceania/Antarctica 27 (224 in total)
- India
- Yes. India is in the Asia list on the Antigravity FAQ (checked 12 October 2026).
- How to apply
- Download Antigravity or install the Antigravity CLI from antigravity.google, then sign in with a personal Google account.
- Watch out for
- No published number for the free agent quota. Heavy agent tasks use up the weekly limit faster.
- Using purchased or promotional AI credits for usage above the limit is only for Google AI Pro/Ultra users. Free users wait for the weekly reset.
- Third-party models are listed for Ultra only.
- Wrapping an Antigravity login inside third-party coding agents or apps, or rotating accounts, breaks the terms and can get the account suspended. Headless mode (agy -p) is meant for your own scripts and CI. For third-party agents Google points to a Gemini Enterprise or Google AI Studio API key.
- Google says usage limits are subject to change.
- Status note
- Google’s developer blog (19 May 2026) says Antigravity CLI is available to everyone. From 18 June 2026 Antigravity replaced Gemini CLI and the Gemini Code Assist IDE extensions for free users and Google AI Pro/Ultra users. The free quota has no published number: the plans page calls it ‘a meaningful quota, refreshed weekly’, with a weekly rate limit, and says the free tier includes all product features such as the CLI. Pro refreshes every five hours up to a higher weekly limit; Ultra refreshes every five hours with the highest limits. Pro and Ultra prices are not shown on the Antigravity pricing page. Business use is billed through Google Cloud, with Standard/Plus from $30 per seat a month or a $0-seat pay-as-you-go option. The pricing page mentions no card step for the $0 plan.
Google Cloud credits with Google AI Pro and Ultra (Google Developer Program premium benefits)
GoogleChanged 27 January 2026Google AI Pro or Ultra subscribers only
Solo developers
$10 a month Google Cloud credit with AI ProBest case: $100 a month with AI Ultra 20x
Google AI Pro and Ultra subscriptions include monthly Google Cloud credits: $10 a month on AI Pro, $40 on AI Ultra 5x and $100 on AI Ultra 20x, usable on Google Cloud and Google Maps Platform products such as Firebase and Vertex AI.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for Google Cloud credits with Google AI Pro and Ultra (Google Developer Program premium benefits)
- Who qualifies
- Active Google AI Pro or Ultra plan linked to your Google Developer Program profile.
- Choose a Cloud billing account on the Developer Program My Benefits page; you need permission to redeem promotions on it (billing.accounts.redeemPromotion).
- Google AI Plus does not include the credits.
- How long
- Monthly while subscribed; each month’s credit expires one year after it is granted.
- Referral or funding needed
- No; needs a paid Google AI Pro or Ultra subscription.
- Card needed
- Yes
- Work or school email
- No
- Countries
- No country limit stated on the benefits FAQ.
- India
- Yes. Google AI plans are sold in India, and the benefits FAQ states no country limit. We did not check the Developer Program’s separate available-regions page. Indian Cloud billing accounts must pass identity verification within 30 days.
- How to apply
- Subscribe to Google AI Pro or Ultra, join the Google Developer Program with the same account, then claim credits on the My Benefits page (or redeem a promo code in the Cloud console).
- Watch out for
- Once applied to a billing account, credits cannot be moved to another one.
- AI Pro and Ultra do not include the old standalone premium extras such as 1:1 consultations, certification vouchers or unlimited Google Skills.
- Google AI Pro costs $19.99 a month in the US.
- Google’s Gemini API billing page (updated 28 September 2026, checked 15 October 2026) says not all Google Cloud credits can pay for the Gemini API and AI Studio, and Prepay users must buy prepaid credit before eligible Cloud credits apply. It does not say whether AI Pro or Ultra credits count. Gemini 4 Argon is not available on the Google AI Pro plan.
- Status note
- Google announced on 27 January 2026 that Developer Program premium benefits are included in AI Pro and Ultra. The standalone premium subscription for personal accounts was wound down: annual auto-renewals ran until 30 March 2026 and monthly until 30 June 2026. Workspace premium was not affected. Ultra is now sold in two tiers (5x and 20x).
Google Colab (free tier)
GoogleActiveNo application; any Google account
StudentsResearchersSolo developersEducatorsAnyone
Free notebooks with GPU/TPU when available
Free browser-based Jupyter notebooks with GPU and TPU access when available. Sessions can run up to 12 hours. GPU types and usage limits are not guaranteed and change with demand.
Eligibility, card, India and how to apply for Google Colab (free tier)
- Who qualifies
- Google account
- Colab’s AI features need a Google account aged 18+ and a supported locale
- Workspace for Education students under 18 need parental consent from their organisation
- How long
- Ongoing; at most 12 hours per notebook session, with idle timeouts
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated in the Colab FAQ
- India
- Yes. The Colab FAQ lists no country restriction. Only the AI features depend on locale and age.
- How to apply
- Open colab.research.google.com, sign in with a Google account and choose a GPU or TPU runtime under Runtime > Change runtime type.
- Watch out for
- No guaranteed GPU. You may get only a CPU runtime at busy times.
- Without a positive compute-unit balance, Colab bans SSH or remote desktop, mainly using a web UI instead of the notebook, chess training and distributed computing workers. Runtimes doing these can be stopped without warning.
- Banned on all runtimes: using several accounts to get around limits, crypto mining, torrenting, file hosting, remote proxies, password cracking, deepfakes and using containers to evade abuse checks.
- Idle runtimes time out, so save work to Drive.
- Status note
- Colab FAQ checked 12 October 2026. Google publishes no fixed free GPU-hour figure: resources are ‘not guaranteed and not unlimited’, and limits change over time. Guaranteed resources are sold through Colab Enterprise or GCP Marketplace. Paid options are Pay As You Go, Pro and Pro+ (Pro+ allows up to 24-hour runs); Google AI plans also include Colab benefits. Colab Pro for Education is closed to new sign-ups. Paid prices sit behind sign-in and were not confirmed. Kaggle gives Colab Pro/Pro+ subscribers 15/30 extra Kaggle GPU hours a week as an experimental, limited-time promotion.
Official page for Google Colab (free tier) (opens in a new tab)Checked 12 October 2026
Grok API free credits (data-sharing credits)
xAI (docs now branded SpaceXAI)Not fully confirmedNo application; credits likely ended
Solo developersBootstrapped startups
Reported data-sharing credits, likely endedBest case: Reported up to $150 a month if opted in
Guides from 2025-2026 describe $25 signup credits and up to $150 a month for teams that opted into data sharing. xAI’s current billing docs mention only prepaid credits and promo codes.
Low confidence: we could not confirm the key details on an official page. Check the official page before you rely on this entry.
Eligibility, card, India and how to apply for Grok API free credits (data-sharing credits)
- Who qualifies
- Reported: opt in to share API inputs and outputs with xAI and have made a prior payment (unconfirmed)
- How long
- Reported monthly while opted in (unconfirmed)
- Referral or funding needed
- No (reported)
- Card needed
- Not stated
- Work or school email
- No
- Countries
- Unknown
- India
- Unclear. xAI’s billing docs say Indian payment cards can buy prepaid credits only through Guest Checkout, so India sign-up works. The free-credit program itself could not be confirmed.
- How to apply
- Check Billing in the xAI Console; do not plan around free credits unless your Console shows them.
- Watch out for
- The reported program required sharing your API data for model training, which rules it out for sensitive data.
- API requests are rejected once prepaid credits run out unless invoiced billing is set up.
- Status note
- xAI’s billing documentation (last updated 29 September 2026) and its docs index (llms.txt) do not mention free or data-sharing credits. Third-party trackers report the $25 signup credit stopped in April 2026 and treat the $150 data-sharing credit as ended; not confirmed officially.
Official page for Grok API free credits (data-sharing credits) (opens in a new tab)Checked 12 October 2026
GroqCloud Free plan
GroqActiveNo application; any GroqCloud account
Solo developersStudentsBootstrapped startupsAnyone
Free tier: 1,000 requests a day on gpt-oss-120b
Free API access to open models on GroqCloud within Free plan rate limits, for example openai/gpt-oss-120b at 30 requests a minute, 1,000 a day, 8K tokens a minute and 200K tokens a day.
Eligibility, card, India and how to apply for GroqCloud Free plan
- Who qualifies
- GroqCloud account
- Limits apply per organisation
- How long
- Ongoing
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated
- India
- Yes. No country limit is stated for the Free plan, and Groq’s Billing FAQ has special billing thresholds for customers with an Indian billing address, so India is served (checked 13 October 2026).
- How to apply
- Sign up at console.groq.com and create an API key.
- Watch out for
- Free plan limits are low; higher limits, Batch and Flex need the paid Developer plan.
- Exact limits for your organisation are on the Limits page and may differ from the summary.
- No payment method is needed for Free; upgrading to Developer needs a credit card, US bank account or SEPA debit account.
- Status note
- Free and Developer plan limit tables listed in GroqCloud docs, checked 13 October 2026 (Free gpt-oss-120b: 30 requests a minute, 1K a day, 8K tokens a minute, 200K a day; Developer base: 1K requests a minute, 250K tokens a minute).
Official page for GroqCloud Free plan (opens in a new tab)Checked 13 October 2026
Inference Providers monthly credits (HF PRO)
Hugging FaceActiveHF PRO ($9 a month) subscribers only
Solo developersResearchersStudents
No free credit; PRO includes $2 a month
Free accounts get no monthly inference credits and must buy credits. HF PRO ($9 a month) includes $2 of compute credits a month for Inference Providers, Inference Endpoints, Spaces hardware and Jobs.
Eligibility, card, India and how to apply for Inference Providers monthly credits (HF PRO)
- Who qualifies
- PRO users: $2 a month.
- Team or Enterprise organisations: $2 per seat a month, shared.
- How long
- Monthly while subscribed to PRO (or Team/Enterprise).
- Referral or funding needed
- No; needs a paid PRO, Team or Enterprise plan for the monthly credit.
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated.
- India
- Yes. No country restriction stated.
- How to apply
- Subscribe to PRO at huggingface.co/subscribe/pro; credits apply automatically to routed requests.
- Watch out for
- Credits apply only to requests routed through Hugging Face, not when you use your own provider key.
- Beyond the credit, usage is pay-as-you-go at the provider’s price with no markup.
- PRO’s ZeroGPU quota is 8x the free quota, but the free quota is not stated.
- Status note
- Hugging Face’s docs, fetched 12 October 2026, list no monthly credits for free users.
Official page for Inference Providers monthly credits (HF PRO) (opens in a new tab)Checked 12 October 2026
Jules (no-cost base plan)
GoogleActiveNo application; 18+, Google account
Solo developersStudentsOpen-source maintainersAnyone
15 agent tasks a day on Gemini 2.5 ProBest case: Up to 300 tasks a day with Google AI Ultra
Google’s asynchronous coding agent works on your GitHub repos. The no-cost plan gives 15 tasks per rolling 24 hours and 3 concurrent tasks on Gemini 2.5 Pro.
Eligibility, card, India and how to apply for Jules (no-cost base plan)
- Who qualifies
- Aged 18 or over
- Google account; Jules connects to GitHub to work on repositories
- Higher limits need Google AI Pro (100 tasks a day, 15 concurrent) or Ultra (300 tasks a day, 60 concurrent), sold only to individual @gmail.com accounts on 12 October 2026
- How long
- Ongoing; task limit is per rolling 24 hours
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated on the usage-limits or FAQ pages
- India
- Unclear. The Jules pages fetched on 12 October 2026 do not list supported countries, and sign-up from India was not tested.
- How to apply
- Sign in at jules.google with a Google account and connect a GitHub account or repositories.
- Watch out for
- Free plan uses Gemini 2.5 Pro; newer models (Gemini 3 Pro onward) are for Pro/Ultra.
- Paid upgrades are only for personal @gmail.com accounts, not Workspace.
- Each member of a shared Google One plan gets their own task limits; limits are not pooled.
- Still labelled public beta, so features and limits can change.
- Status note
- Usage-limits page checked 12 October 2026. The FAQ still calls Jules a public beta and says one plan is available to all users without cost. Pro and Ultra get newer models, starting with Gemini 3 Pro. Business users can fill in an interest form; upgrade paths for other account types are still in development. Google says limits may change. Neither page mentions Antigravity or any shutdown.
Official page for Jules (no-cost base plan) (opens in a new tab)Checked 12 October 2026
Kaggle Notebooks free GPU and TPU quota
Kaggle (Google)ActiveNo application; any Kaggle account
StudentsResearchersSolo developersAnyone
30 GPU hours + 20 TPU hours a week
Free Kaggle notebooks with a weekly quota of 30 GPU hours (sometimes more) and 20 TPU hours. Sessions run up to 12 hours (9 for TPU). GPU options are one P100 or two T4s; the TPU is a v3-8.
Eligibility, card, India and how to apply for Kaggle Notebooks free GPU and TPU quota
- Who qualifies
- Kaggle account
- Phone verification for accelerators and internet access is widely reported but was not confirmed on the docs pages checked
- How long
- Ongoing; quota resets weekly
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated in Kaggle docs
- India
- Yes. Kaggle’s docs list no country restriction.
- How to apply
- Create a Kaggle account, open a notebook and pick a GPU or TPU under Settings > Accelerator.
- Watch out for
- Interactive sessions end after 20 minutes idle. Use ‘Save & Run All’ for long jobs, which must finish within 12 hours (9 for TPU).
- You may wait in a queue for a GPU or TPU machine at busy times.
- Only /kaggle/working (20 GB) is saved; scratch disk is lost after the session.
- Kaggle can change the quotas and the Colab-linked bonus at any time.
- Status note
- Kaggle docs checked 12 October 2026. GPU quota: 30 hours a week ‘or sometimes higher depending on demand and resources’. TPU v3-8: up to 20 hours a week, 9 hours per session. Each session gets 20 GB of auto-saved disk; CPU sessions get 30 GB RAM, GPU sessions 29 GB, the TPU VM 330 GB. Colab Pro and Pro+ subscribers who link accounts get 15 and 30 extra GPU hours a week; Kaggle labels this experimental and a limited-time promotion it may change without notice.
Official page for Kaggle Notebooks free GPU and TPU quota (opens in a new tab)Checked 12 October 2026
Kiro Free plan
Amazon Web Services (Kiro)ActiveNo application; Google/GitHub/AWS login
Solo developersStudentsAnyone
50 credits a month incl. Claude Sonnet 4.5
A permanent free tier of AWS’s agent-based IDE and CLI: 50 credits a month on open-weight models and Claude Sonnet 4.5, with rate limits. New users get $20 off their first paid upgrade.
Eligibility, card, India and how to apply for Kiro Free plan
- Who qualifies
- Sign in with Google, GitHub or AWS Builder ID
- No add-on credits on Free; a paid plan is needed to buy more
- How long
- Ongoing; the 50-credit allowance renews each billing cycle
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated on pricing or billing pages
- India
- Yes. No country restriction is stated, and sign-in uses Google, GitHub or Builder ID accounts. Paid upgrades need a credit card.
- How to apply
- Download Kiro from kiro.dev and sign in. New accounts start on the Free tier.
- Watch out for
- 50 credits run out fast. Agent tasks use different amounts of credit per prompt.
- The $20 first-upgrade credit only applies to social or Builder ID logins, not AWS IAM Identity Center. It is prorated by the days left in the month.
- Paid subscriptions are charged on the 1st of each calendar month and need a credit card.
- The pages do not say unused free credits carry over; treat the allowance as use-it-or-lose-it each cycle.
- Status note
- Pricing and billing docs checked 12 October 2026; the billing doc was last updated 4 August 2026 and calls Free ‘perpetual’. Free models named on the pricing page: open-weight models such as Qwen3 Coder Next, DeepSeek 3.2 and MiniMax M2.1, plus Claude Sonnet 4.5. Paid plans: Pro $20 (1,000 credits), Pro+ $40 (2,000), Pro Max $100 (5,000), Power $200 (10,000). Add-on credits on paid plans cost $0.04 each (packs $5 for 125 up to $100) and expire 12 months after purchase. GovCloud has no Free tier. AWS names Kiro as the replacement for the Amazon Q Developer IDE plugins, which stopped new sign-ups on 15 May 2026.
Official page for Kiro Free plan (opens in a new tab)Checked 12 October 2026
Mistral API Free mode
Mistral AIActiveNo application; no card needed
Solo developersStudentsAnyone
Free API keys at the lowest rate limits
Free API access in Mistral Studio: keys work by default with no credit card, within the lowest rate limits and the monthly usage included with your plan.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for Mistral API Free mode
- Who qualifies
- Anyone with a Mistral account.
- Meant for evaluation and prototyping.
- How long
- Ongoing.
- Referral or funding needed
- No
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated on the pages checked.
- India
- Yes. No country restriction stated on the pages checked.
- How to apply
- Sign up at console.mistral.ai (Studio) and create an API key.
- Watch out for
- Mistral lets you stop API data being used to improve its services with the ‘Anonymous improvement data’ toggle under Privacy in the Admin panel. Check it before sending sensitive data.
- Buying prepaid credits does not raise rate limits; only billed usage does.
- Status note
- Mistral’s docs call it Free mode. Exact limits appear only on the Limits page in the Admin panel. With pay-as-you-go on, tiers rise with billed usage: Tier 2 above $20 billed, Tier 3 above $100, Tier 4 above $500.
Official page for Mistral API Free mode (opens in a new tab)Checked 12 October 2026
Monthly API credits for Claude Max and Team plans
AnthropicActiveClaude Max or Team subscribers only
Solo developersBootstrapped startupsFunded startupsNonprofitsResearchers
$100 a month API credit on Max 5xBest case: $200 on Max 20x; Team pools up to $500 a month
Claude Max and Team plans include a monthly Claude API credit: $100 on Max 5x, $200 on Max 20x, and $20 per Team Standard seat or $100 per Premium seat, pooled and capped at $500 a month.
Eligibility, card, India and how to apply for Monthly API credits for Claude Max and Team plans
- Who qualifies
- Max 5x, Max 20x, Team, and discounted Team plans (Nonprofit, Scientists).
- Free, Pro and Enterprise plans are not eligible.
- Max: you must be the subscriber. Team: Primary Owner or Owner.
- You link one Claude Console organization where you hold the Owner, Admin or Billing role. Claim on claude.ai in a web browser.
- How long
- Monthly while subscribed; unused credit expires at the end of each billing cycle.
- Referral or funding needed
- No; needs a paid Max or Team plan, held for at least 7 days.
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated in the help article.
- India
- Yes. No country restriction stated. No payment method is needed on the Claude Platform to claim or use the credit.
- How to apply
- On claude.ai open Settings > Billing (Team: Organization settings > Billing), choose Link organization, pick or create the Console organization and accept the Supplemental Credit Terms.
- Watch out for
- You can link only one Console organization; only support can change it.
- Covers the Claude API, Console Playground, Claude Managed Agents and the Agent SDK. Not usable for interactive Claude Code or extra usage in Claude apps.
- Not usable for Claude on Amazon Bedrock, Google Vertex AI or Microsoft Foundry.
- Every API key in the linked organization draws on the same balance. Without purchased credits, API requests stop when it runs out.
- Status note
- New in October 2026; the help centre says it is rolling out over a few days and gives no start date (our Claude guide recorded it from about 7 October). This credit is separate from the Claude for Startups $1,000 credit, which Anthropic paused on 9 October 2026. On 15 October 2026 the whole Claude for Startups program stopped taking new applications (current cohort full).
NVIDIA API Catalog free endpoints (build.nvidia.com) via NVIDIA Developer Program
NVIDIAActiveNo application; free NVIDIA dev account
Solo developersStudentsResearchersBootstrapped startupsAnyone
Free rate-limited API on 38 open-model endpoints
Free, rate-limited API calls to models tagged ‘Free Endpoint’ on build.nvidia.com (38 of 97 catalogue models on 12 October 2026), mostly open models, plus NIM container downloads for development and testing.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for NVIDIA API Catalog free endpoints (build.nvidia.com) via NVIDIA Developer Program
- Who qualifies
- Free NVIDIA Developer Program / build.nvidia.com account (model pages require login for API use)
- Meant for evaluation, prototyping, development and testing by individuals. NVIDIA directs production use to NVIDIA AI Enterprise or paid partner endpoints.
- How long
- Ongoing; each model’s free endpoint can be deprecated
- Referral or funding needed
- No
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated on the pages checked
- India
- Unclear. No country list was found on NVIDIA’s pages. Sign-up from India was not confirmed.
- How to apply
- Sign in at build.nvidia.com with an NVIDIA account, pick a model tagged Free Endpoint and generate an API key.
- Watch out for
- Not for production. Free endpoints can be throttled under load and retired per model.
- Community forum threads report that the 40 RPM limit is not raised on request. Dedicated capacity needs your own NIM deployment or a paid partner endpoint.
- Most catalogue models are downloadable or partner-hosted only, not free endpoints.
- Exact terms of service for free use were not confirmed on an official page.
- Status note
- Checked 12 October 2026: the models catalogue lists 97 models; its ‘Free Endpoint’ filter shows 38. The site’s page data gives ‘Up to 40 rpm’ and ‘10,000 requests per day’, and says limits vary by model and traffic from other users can cause throttling. These figures come from the site’s embedded copy text, not a rendered FAQ (build.nvidia.com/faq returns 404). The NIM page offers free access to NIM API endpoints for prototyping. An NVIDIA forum FAQ by an NVIDIA account (September 2024) calls the catalogue a trial experience with limits meant for evaluation and says the Developer Program is for individuals doing development, research and testing. Example of retirement: the Llama 3.1 8B Instruct free endpoint was marked for deprecation on 25 August 2026. Third parties say an older credit system was replaced by rate limits; this was not confirmed on an official page.
OpenRouter free models (:free variants)
OpenRouterActiveNo application; any OpenRouter account
Solo developersStudentsAnyone
Free models: 20 requests a minute, 50 a dayBest case: 1,000 a day after buying 10 credits
Free ‘:free’ model variants: 20 requests a minute and 50 a day, rising to 1,000 a day once you have bought at least 10 credits in total.
Eligibility, card, India and how to apply for OpenRouter free models (:free variants)
- Who qualifies
- Any OpenRouter account.
- The higher daily limit starts at 9 credits purchased, to absorb fees.
- How long
- Ongoing.
- Referral or funding needed
- No. A one-off purchase of 10 credits raises the daily limit.
- Card needed
- No
- Work or school email
- No
- Countries
- No country limit stated.
- India
- Yes. No country restriction stated. Credits can be paid by card, AliPay or USDC.
- How to apply
- Sign up at openrouter.ai, create an API key and call a model ID ending in ‘:free’.
- Watch out for
- Buying credits by card (Stripe) costs a 5.5% fee ($0.80 minimum); crypto costs 5%.
- OpenRouter’s terms let it expire unused credits one year after purchase.
- Providers that log prompts, or whose policy OpenRouter cannot confirm, are used only if you turn on the model-training toggle in privacy settings.
- Status note
- The list of free models changes often.
Official page for OpenRouter free models (:free variants) (opens in a new tab)Checked 12 October 2026
Ended and paused programs (3)
Amazon Q Developer Free tier
Amazon Web ServicesEndedClosed to new sign-ups 15 May 2026
Solo developersStudentsAnyone
Ended: was 50 agent requests a month free
A free tier of AWS’s AI coding assistant in the IDE and CLI: 50 agent requests and 1,000 lines of Java upgrade transformation a month. It closed to new sign-ups on 15 May 2026; Kiro replaces it.
Eligibility, card, India and how to apply for Amazon Q Developer Free tier
- Who qualifies
- Was available through AWS Builder ID in the IDE plugins; new Free Tier account creation is now blocked
- How long
- Closed to new sign-ups 15 May 2026; IDE plugins reach end of support 30 April 2027
- Referral or funding needed
- No (when it was open)
- Card needed
- Not applicable
- Work or school email
- Not applicable
- Countries
- Closed to new sign-ups worldwide
- India
- No. New sign-ups are blocked worldwide. Use Kiro Free instead.
- How to apply
- No new sign-ups. Use Kiro (kiro.dev) instead.
- Watch out for
- The pricing page has not been updated and still advertises the free tier.
- Java upgrades and .NET porting in the IDE plugins end on 30 April 2027; AWS points to AWS Transform as the alternative.
- AWS says the newest coding models are only on Kiro.
- Status note
- AWS announced the end of support on 30 April 2026. From 15 May 2026, new Q Developer Free Tier accounts (via Builder ID in the IDE plugins) and new subscriptions (via the AWS Console) are blocked. Opus 4.6 was removed from Q Developer Pro on 29 May 2026. The IDE plugins and paid subscriptions reach end of support on 30 April 2027. AWS names only Pro and Kiro subscribers as keeping plugin access until then; it says nothing direct about existing free users. Q Developer in the AWS Console, the AWS documentation and marketing sites, the Console mobile app, and Slack/Teams is not affected. The pricing page still shows the free-tier limits and Pro at $19 per user a month, with no mention of the sunset.
Official page for Amazon Q Developer Free tier (opens in a new tab)Checked 12 October 2026
Gemini Code Assist for individuals and free Gemini CLI (Login with Google)
GoogleEndedEnded 18 June 2026
Solo developersStudentsAnyone
Ended: was 1,000 model requests a day free
A free Google-login tier for Gemini CLI and the Gemini Code Assist IDE extensions, with up to 1,000 model requests a day. It stopped serving requests on 18 June 2026, and Google now points users to Antigravity.
Eligibility, card, India and how to apply for Gemini Code Assist for individuals and free Gemini CLI (Login with Google)
- Who qualifies
- Was open to personal Google accounts
- Gemini CLI now works with Gemini Code Assist Standard/Enterprise licences or paid Gemini API keys
- How long
- Ended 18 June 2026
- Referral or funding needed
- No (when it ran)
- Card needed
- Not applicable
- Work or school email
- Not applicable
- Countries
- Ended for all countries
- India
- No. The free consumer tier ended worldwide on 18 June 2026. Indian users can use the free Antigravity plan instead.
- How to apply
- No longer available. Use the free Antigravity plan, a Gemini API key, or a Gemini Code Assist Standard/Enterprise licence.
- Watch out for
- Older guides still quote 60 requests a minute and 1,000 a day for free Gemini CLI. That tier is gone.
- The Gemini CLI repo’s quota page was not updated, so do not rely on it for the consumer tier.
- Standard and Enterprise licences ($19-22.80 and $45-54 per user a month) are not permanent either; Google says they have their own deprecation schedule.
- Status note
- Google announced the change on 19 May 2026. On 18 June 2026 Gemini CLI and the Code Assist IDE extensions stopped serving requests for Gemini Code Assist for individuals, Google AI Pro and Google AI Ultra users, and ‘Login with Google’ no longer works for them. The Code Assist homepage banner confirms the replacement by Antigravity on that date. Standard (1,500 agent/CLI requests per user a day) and Enterprise (2,000) keep access for now, but the deprecation page (updated 9 October 2026) says they are on a separate deprecation schedule. Google’s blog says Gemini CLI stays usable with paid Gemini API keys; it does not address free API keys. The Gemini CLI repo’s quota doc still lists a 1,000 a day Google-login tier and a 250 a day Flash-only free API-key tier, which predate the switchover. The consumer Gemini Code Assist GitHub app stopped accepting new installations on 18 June 2026.
GitHub Models (free model playground and inference API)
GitHubEndedRetired 30 July 2026
Solo developersAnyone
Retired: was a free model playground and API
Ended. GitHub Models gave developers a model playground, model catalog and inference API inside GitHub. It was fully retired on 30 July 2026 and is no longer available to anyone.
Eligibility, card, India and how to apply for GitHub Models (free model playground and inference API)
- Who qualifies
- Not available to anyone since 30 July 2026
- How long
- Ended 30 July 2026
- Referral or funding needed
- Not applicable (retired)
- Card needed
- Not applicable
- Work or school email
- Not applicable
- Countries
- Retired worldwide
- India
- No. Retired for every user worldwide. Indian developers should use one of the free AI API alternatives instead.
- How to apply
- You can no longer sign up. GitHub points users to Azure AI Foundry (called Microsoft Foundry in the changelog) and GitHub Copilot. For free AI APIs, see our Gemini API free tier, Groq free plan, OpenRouter free models and Cerebras entries.
- Watch out for
- Old tutorials and blog posts still describe GitHub Models as a free AI API. It no longer works.
- GitHub Spark apps that called llm() stopped working on 30 July 2026; they need a separate inference provider with your own API key and billing.
- GitHub’s suggested replacement, Azure AI Foundry, is an Azure service; GitHub’s pages do not describe a free tier for it.
- Status note
- 16 June 2026: closed to new organisations and enterprises that had never used it, on free and paid plans. 1 July 2026: full retirement announced, with brownouts on 16 and 23 July 30 July 2026: playground, model catalog, inference API and BYOK removed for all customers, including those with active usage. GitHub Spark, which used GitHub Models for its llm() function, stopped taking new users on 4 August 2026 and closed to existing users after 31 August 2026. The old free-tier rate limits are no longer documented by GitHub.
Does OpenAI give free API credits?
No free credit at sign-up. On 12 and 13 October 2026 the OpenAI API needed a positive account balance, and its startup credits needed a referral code from a partner VC. These routes do give something free:
- Complimentary daily tokens. Eligible organisations that share API inputs and outputs with OpenAI get free tokens every day. The Build tier gets 250K tokens on flagship models and 2.5M on mini models; the Launch and Grow tiers get 1M and 10M. The allowance resets at 00:00 UTC. Shared data can be used to train OpenAI’s models.
- OpenAI for Startups. Credits, rate-limit upgrades and time with OpenAI engineers, but only with a referral code from a VC in OpenAI’s partner network. OpenAI does not publish amounts. Other startups get Build Hours, starter repos and events. OpenAI also hands out credits at some events and hackathons, where it asks for your organisation ID.
- Researcher Access Program. Up to $1,000 in API credits for research on responsible AI deployment, safety, and AI’s social and economic impact. Applications are reviewed in March, June, September and December, credits arrive 4 to 6 weeks after the decision, and they expire after a year.
- Codex credits for university students. $100 of Codex usage for university students in the US and Canada. These are ChatGPT credits for Codex, not API credits.
- Codex for Open Source (medium confidence). Six months of ChatGPT Pro with Codex for core maintainers of widely used projects. API credits are possible case by case; OpenAI does not publish the amount per project.
How the daily tokens work
An organisation owner turns on sharing in the API platform’s data-sharing settings. The offer appears there only if OpenAI has made your organisation eligible. The catches:
- Only traffic on projects with sharing turned on counts, and the quota is shared across all models in each group.
- If one request would push the day’s total over the quota, the whole request is billed.
- Fine-tuned models, fine-tuning training, evals and tool use are excluded.
- Organisations with Zero Data Retention, and Enterprise customers, cannot opt in.
OpenAI says it will give 30 days’ notice before ending the offer. Turn it on only for projects that never touch confidential data. Using Codex for finance work? See our guide to Codex for finance teams.
Does Claude give free API credits?
No free credit at sign-up. On 12 and 13 October 2026 the Claude API credit most people could get came with a paid Max or Team plan. These are the routes we track:
- Monthly API credit with Claude Max and Team. $100 a month on Max 5x and $200 on Max 20x. Team plans get $20 per Standard seat or $100 per Premium seat, pooled and capped at $500 a month. You must have held the plan for 7 days. The credit pays for the Claude API, Console Playground, Claude Managed Agents and the Agent SDK, but not for interactive Claude Code, and not for Claude on Amazon Bedrock, Google Vertex AI or Microsoft Foundry. Unused credit expires at the end of each billing cycle. See how to claim the monthly API credits.
- Claude for Startups (Claude Startups program). Not taking new applications: on 15 October 2026 the program page said the current cohort is full, gave no reopening date and pointed founders to its newsletter. The one-time $1,000 API credit announced on 6 October 2026 was paused on 9 October 2026 and is no longer offered to new startups. Existing members keep Startup Stack partner offers worth up to $45,000. The page still says startups backed by a VC in Anthropic’s partner network may get up to $100,000 in additional API credits, claimed through that VC. Read our Claude for Startups guide.
- External Researcher Access Program. $1,000 in API credits (more in rare cases) for AI safety and alignment research.
- AI for Science. Up to $20,000 in API credits for 6 months, for researchers at academic and nonprofit institutions, with a focus on biology and life sciences.
- Claude for Open Source. Six months of Max 20x for qualifying maintainers. Max plans carry the monthly API credit, but the program terms do not say whether complimentary plans get it.
Claiming the Max or Team credit. Claim it on claude.ai in a web browser (Max: Settings, then Billing; Team owners: Organization settings, then Billing) by linking one Claude Console organisation; only support can change the link later. Every API key in that organisation draws on the same balance, and without purchased credits, API requests stop when it runs out.
Use our AI leaderboard to compare models before you spend free credits, and our Claude Code tutorial for the coding side.
Is the Gemini API free?
Yes, with limits: the Gemini API free tier covers Gemini 3.x Flash and Flash-Lite, Gemini 2.5 Pro and Flash, Gemma 4 and embeddings, with no card (checked 12 and 13 October 2026). Pro previews and image or video models are paid-only. Google shows the exact free limits only inside AI Studio, and daily request quotas reset at midnight Pacific time.
Gemini 4 Argon is not free. Google’s pricing page, updated 14 October 2026, says Argon, its new flagship model, is available on neither the Free tier nor the Google AI Pro plan (checked 15 October 2026). Our Gemini 4 Argon guide covers its pricing and access.
Who can use it. You must be 18 or over and in an available region. India, the UK and every EU country are on the list. You need a Google Cloud project, but no billing account.
What Google does with free-tier data. Free-tier prompts and outputs are used to improve Google products and may be read by human reviewers. In the EEA, Switzerland and the UK, the paid-service data terms apply even to free use, and only paid services may be used in apps for users there.
What the free tier leaves out:
- Google Search grounding (answers backed by live Google Search results) on Gemini 3 and newer models. The 5,000 free grounding requests a month apply to paid use.
- Batch and Flex, cheaper ways to send requests that return results more slowly, are generally not on the free tier.
- The Google Cloud $300 trial credit cannot pay for Gemini API use in AI Studio.
When you move to paid. New paid users start on Prepay, with a $5 minimum top-up. Paid prices for Gemini 3.8 Flash and 3.6 Flash double from 1 January 2027, so check the pricing page before you budget for next year.
Google AI Pro and Ultra credits. If you pay for Google AI Pro or Ultra, your plan includes $10, $40 or $100 a month in Google Cloud credits, depending on the plan (medium confidence). Google says they can be spent on any Google Cloud or Google Maps Platform product, including Vertex AI, and each month’s credit expires a year after it is granted. Do not count on them for your Gemini API bill. Google’s Gemini API billing page says not all Cloud credits can pay for Gemini API use, and that Prepay users must buy prepaid credit before eligible Cloud credits apply. It does not say whether AI Pro or Ultra credits are eligible (checked 15 October 2026).
AI API credits for startups: which credits pay for which models
A startup credit is only useful if it pays for the model you call. Cloud credits usually pay for the cloud’s own services, and several exclude third-party models, so a large cloud credit can leave your Claude or OpenAI bill untouched. We filled this table only from each program’s terms; where the terms are silent, it says so.
| Credit program | Pays for these models | Does not pay for | How you get it |
|---|---|---|---|
| AWS Activate (Activate Credits) | AWS usage, including third-party models on Amazon Bedrock: the one AWS Marketplace exception the credits cover | Other AWS Marketplace purchases | Founders tier starts at $1,000 with no investor |
| Google for Startups Cloud Program (Start and Scale tiers, AI-first) | Google Cloud usage. Specific models: not stated | Third-party models and Google Cloud Marketplace offerings, which are billed directly | Start tier: $2,000 with no investor |
| Microsoft for Startups (Azure Startup credits) (medium confidence) | Azure usage. Specific models: not stated | Not stated | $200 on joining, then milestones |
| Cloudflare for Startups | Workers AI, capped at $2,500, $10,000 or $50,000 by tier | AI Gateway (temporarily not covered on 12 and 13 October 2026) | Tier 3: $10,000 with no investor |
| Claude for Startups (Claude Startups program) | Claude on the first-party Claude API only | Claude on AWS Bedrock, Google Vertex AI or other third-party platforms | No new applications on 15 October 2026 (cohort full); up to $100,000 only through a partner VC |
| Claude Max and Team credits | Claude API, Console Playground, Claude Managed Agents and the Agent SDK | Bedrock, Vertex AI and Microsoft Foundry; interactive Claude Code | Included with a paid Max or Team plan |
| DigitalOcean Startups (replaced Hatch) | Core DigitalOcean usage. Specific models: not stated | Since 13 May 2026: third-party models such as Anthropic and OpenAI, Serverless and Dedicated Inference, and GPU products | Amount set through partners |
| Vercel for Startups | AI Gateway from the Flexible Commitment, at most 50% of each month’s credit | Gateway use above that 50%, which is billed even if other credit is unused | Only through an approved partner |
| OpenAI for Startups | OpenAI API | Not stated | Partner VC referral code only; amounts not published |
| Alibaba Cloud AI Catalyst Program | Alibaba Cloud credits plus up to 2 billion free Model Studio tokens | Marketplace, domains, CDN and prepaid products | AI-first startups; campaign runs to 31 March 2027 |
How to use it. If you build on Claude, AWS Activate is the one cloud credit here whose terms cover third-party models on Amazon Bedrock, which is one place Claude runs. Claude for Startups credits and the Max and Team credits work the other way round: only on Anthropic’s own API, never on Bedrock. If you build on OpenAI, the only direct startup route is a partner VC. You can hold several of these at once, but each bill draws only on its own provider’s credit: a Bedrock bill uses AWS credits, and a Claude Console bill uses Anthropic credits. Adding the amounts up tells you nothing about what one model will cost you.
When AI credits run out
- OpenAI for Startups: usage switches to standard billing automatically.
- Modal: a payment method is required so apps keep running after the credits end.
- Baseten: credits expire 6 months after they are granted, shorter than most programs.
- Vercel for Startups: after a year, or once the amount is used, the account moves to standard terms, billed monthly at list rates.
- Cerebras trial: access pauses; you are not charged or moved to paid use automatically.
Credits are not cash, and some programs say you are responsible for any tax on their value. Ask your accountant how to treat them. For every startup cloud program, see free credits for startups. For cloud free trials and always-free tiers, see free cloud credits. Our Cloudflare for Startups guide shows how to apply, step by step.
AI startup programs and GPU credits
AI model, GPU and inference programs for startups, plus Hugging Face’s GPU grants for demos. Programs that state what a typical applicant without a referral gets come first, largest first. Programs that give nothing without a referral or publish only a maximum follow, largest maximum first. Programs that publish no amount come next, A to Z, and entries we could not fully confirm come last. Research credits from OpenAI, Anthropic and Lambda are on our credits page for researchers. Each entry says whether you need a referral, a funding stage or a card, checked 12 and 13 October 2026. Use No referral needed to hide programs that only partner-backed startups can join. Low-confidence entries carry a caution line: do not plan around them until the provider confirms.
Who it is for
Show only
Status
No programs match these filters.
Skip past this listTogether AI Startup Accelerator
Together AIActiveNo referral; selection-based
Bootstrapped startupsFunded startupsAccelerator startups
Up to $15,000 credits (Build tier, up to $5M raised)Best case: Up to $50,000 if you raised over $10M
Platform credits for serverless inference, dedicated endpoints, fine-tuning and instant clusters, plus forward-deployed engineering hours: up to $15K (Build), $30K (Scale) or $50K (Grow).
Eligibility, card, India and how to apply for Together AI Startup Accelerator
- Who qualifies
- Selection-based program
- Build tier: raised up to $5M (3 hours of engineering time)
- Scale tier: $5M-$10M raised (6 hours)
- Grow tier: over $10M raised (10 hours)
- How long
- Not stated
- Referral or funding needed
- No referral; tier depends on money raised (Build up to $5M, Scale $5M-$10M, Grow over $10M)
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated (page metadata says it serves startups worldwide)
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Submit the application form on the Startup Accelerator page.
- Watch out for
- Credits do not apply to Reserved GPU Clusters.
- Selection-based, so meeting a tier’s funding range does not guarantee entry or the maximum amount.
- Credit expiry is not published.
- Status note
- Amounts are ‘up to’ figures per tier, checked 12 October 2026. Also includes go-to-market support and VC introductions.
Official page for Together AI Startup Accelerator (opens in a new tab)Checked 12 October 2026
Rudra startup program (Shakti Cloud credits)
Yotta Data ServicesActiveAI/ML startups, bootstrapped to Series A
Bootstrapped startupsFunded startups
$10,000 GPU credits at Launch stage (first tranche $1K)Best case: Up to $100,000 at the Thrive stage
Shakti Cloud GPU credits by stage: Launch $10,000 (6 months), Ignite $25,000 (6 months), Fuel $50,000 (9 months), Thrive $100,000 (12 months), paid in tranches, plus 25% off extra usage and mentoring.
Eligibility, card, India and how to apply for Rudra startup program (Shakti Cloud credits)
- Who qualifies
- Working on AI/ML, data science or high-performance computing
- Active company website and LinkedIn pages (company and applicant)
- Bootstrapped or funded up to Pre-Seed, Angel, Seed, debt or Series A
- Stage set by age and revenue/funding, e.g. Launch: under 1 year and under $50K; Thrive: under 10 years and under $2.5M revenue or over $1M funding
- No earlier Yotta credits of equal or greater value
- How long
- 6 to 12 months depending on stage, from entry into that stage
- Referral or funding needed
- No: bootstrapped up to Series A is accepted. Referrals from Yotta’s funding partners are welcome but not required
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated; Shakti Cloud runs in Indian data centres
- India
- Yes. Indian GPU cloud and an IndiaAI compute provider. You must complete KYC and sign Yotta’s agreements.
- How to apply
- Use the apply-now form on yotta.com/rudra with a business plan, pitch deck, team bios, financial projections and KYC. Yotta confirms acceptance within 3 business days of finishing its evaluation.
- Watch out for
- Credits arrive in tranches, and the first is only $1,000. The rest comes as you qualify, despite the homepage line about up to $35K immediately.
- The Ignite tranches listed ($1,000 + $4,000 + $15,000) add up to $20,000, not the stated $25,000. Ask Yotta to confirm.
- If you enter at a later stage, you do not get the earlier stages’ credits; the overall cap is $100,000.
- Unused credits lapse when the stage’s validity ends, and you pay monthly invoices for usage beyond credits.
- You sign a program agreement, an MSA and service order forms, and Yotta gets a licence to use your name and logo in its marketing. Yotta can close the credit account or discontinue the program with notice.
- Status note
- Rudra pages carry a 2026 copyright. The eligibility page no longer lists NVIDIA Inception membership. yotta.com/shambho now redirects to Rudra, but the Rudra terms page still uses ‘Shambho’ wording.
Official page for Rudra startup program (Shakti Cloud credits) (opens in a new tab)Checked 12 October 2026
E2E Networks Startup Program (AI Explorer, AI Builder, AI Scaler)
E2E NetworksActiveAI startups <5 yrs old, max 10 staff
Bootstrapped startupsFunded startupsAccelerator startups
$5,000 free GPU credits for 12 months (AI Explorer)Best case: Up to $100K usage discounts with $5M+ raised
Indian GPU cloud. AI Explorer: $5K (Rs 5 lakh) free credits for small early-stage AI startups. AI Builder: up to $50K usage discounts ($2M+ raised). AI Scaler: up to $100K usage discounts ($5M+ raised).
Eligibility, card, India and how to apply for E2E Networks Startup Program (AI Explorer, AI Builder, AI Scaler)
- Who qualifies
- Legally incorporated company in any country
- AI/ML startup with GPU compute needs
- AI Explorer: 10 or fewer employees, founded within the last 5 years
- AI Builder: founded within 5 years; AI Scaler: founded within 7 years
- Excludes established enterprises or corporate subsidiaries, consultancies or agencies, crypto mining and credit resale
- How long
- 12 months from activation; extensions possible on request
- Referral or funding needed
- No for AI Explorer (VC or accelerator backing preferred, not required). AI Builder needs $2M+ raised and AI Scaler $5M+
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit (startups incorporated anywhere); capacity is in India
- India
- Yes. Indian provider and an empanelled IndiaAI compute provider. Tiers are also shown in INR.
- How to apply
- Fill in the 4-step form on e2enetworks.com/startup-program (your details incl. LinkedIn, company info, GPU needs, submit).
- Watch out for
- Only AI Explorer gives free credits. The Builder and Scaler benefits are usage-based discounts.
- The headline ‘$100K in credits’ is a discount ceiling, not free money.
- Credits expire 12 months after activation, then pay-as-you-go applies.
- Low-latency serving outside Asia may not suit you, because the capacity is in India.
- Status note
- Live page as of 12 October 2026. Applications are reviewed in 3 to 5 business days. E2E’s older MeitY-startup credit page returns 404.
Deepgram Startup Program
DeepgramActiveOpen: no referral needed
Bootstrapped startupsFunded startupsAccelerator startups
$1,000 in voice API credits to startBest case: Up to $100,000 with usage growth or a partner VC
Speech-to-text, text-to-speech and Voice Agent API credits. Direct applicants start at $1,000; tiers go up to $100,000 over 12 months as usage grows or through partner VCs and accelerators.
Eligibility, card, India and how to apply for Deepgram Startup Program
- Who qualifies
- AI builder or early-stage startup; raises above $10M judged case by case
- In production or launching within 6 months
- Building its own voice AI product; agencies and consultants excluded
- Active Deepgram account and Project ID
- Existing Deepgram customers can apply
- How long
- Each credit grant expires 12 months after issue
- Referral or funding needed
- No; partner VCs, accelerators and startup networks get larger offers listed on their perks pages
- Card needed
- Not stated
- Work or school email
- Yes
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated in the page or terms; we did not confirm India sign-up on an official page.
- How to apply
- Create a Deepgram account, note your Project ID and apply on the startup program page with a work email.
- Watch out for
- Program credits can be used only after the $200 signup credit is used up.
- Credits work only on the Pay As You Go plan, at Pay As You Go prices.
- Credits are tied to the project, not transferable and have no cash value.
- The terms give Deepgram a perpetual licence to use your company name and logo in its marketing.
- Status note
- Checked 12 October 2026. Responses usually within one week; credits arrive in the console 24 to 48 hours after approval.
Official page for Deepgram Startup Program (opens in a new tab)Checked 12 October 2026
Runpod Startup Program
RunpodActiveSelective; needs a working product
Funded startupsAccelerator startups
$1,000 in GPU credits (Starter tier)Best case: $25,000 bonus on a $50,000 upfront commitment
Starter tier: $1,000 in GPU credits. Growth tier: commit $50,000 upfront and Runpod adds $25,000 in bonus credits ($75,000 total) under a 12-month agreement.
Eligibility, card, India and how to apply for Runpod Startup Program
- Who qualifies
- Selective; mainly new Runpod customers
- Starter: pre-Series A with a working product and near-term GPU needs
- Growth: Seed to Series B+ spending or ready to spend $5K-$25K+ a month on GPUs
- Ready to onboard now
- How long
- Growth tier: 12-month agreement; Starter duration not stated
- Referral or funding needed
- No referral, but venture backing is strongly preferred. The Growth bonus requires a $50,000 upfront commitment.
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Apply on the startup program page; if not accepted you can still use Runpod pay-as-you-go.
- Watch out for
- The $25,000 is a match on a $50,000 deposit, not free money for everyone.
- The Growth bonus is one-time with no reload.
- Bootstrapped and pre-funding teams are usually pointed to self-serve pay-as-you-go.
- Status note
- Two-tier structure checked 12 October 2026. Runpod also links separate Creator and academic research offers.
Official page for Runpod Startup Program (opens in a new tab)Checked 12 October 2026
Baseten AI Startup Program
BasetenActiveSeed to Series A, under 5 years old
Funded startupsAccelerator startups
Up to $25,000 inference credits + $2,500 for Model APIs
Up to $25,000 in credits for dedicated inference or training, plus up to $2,500 for Model APIs. Also faster support, event invites and launch/GTM help.
Eligibility, card, India and how to apply for Baseten AI Startup Program
- Who qualifies
- AI must be core to your main products
- Seed to Series A, founded less than 5 years ago
- Net new customer who has never received Baseten credits
- Your company domain must not have had a paid Baseten plan within 31 days of applying
- How long
- Credits should be used within 6 months of the grant date; unused credits then expire
- Referral or funding needed
- No referral code needed, but you must be Seed to Series A. Bootstrapped or pre-seed teams do not fit the stated rules. Amounts are ‘up to’ and acceptance is at Baseten’s sole discretion.
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Yes. The program page and legal terms name no country exclusions. Indian AI startups at Seed to Series A can apply.
- How to apply
- Use the application link on baseten.co/startup-program (form at baseten.co/talk-to-us/startup-program). Baseten decides at its discretion.
- Watch out for
- Credits expire 6 months after they are granted, shorter than most startup programs.
- Excluded: IPO’d or acquired companies, educational institutions, government bodies, nonprofits, personal blogs, dev shops, consultancies, agencies, crypto miners and companies distributing tokens against regulatory guidance.
- The Model API credit is capped separately at $2,500. Most of the value is for dedicated deployments or training.
- You lose eligibility if your domain had a paid plan within 31 days before you apply.
- Status note
- Amounts, eligibility and the 6-month expiry re-confirmed on Baseten’s program page and legal disclaimers on 12 October 2026.
Official page for Baseten AI Startup Program (opens in a new tab)Checked 12 October 2026
LangSmith for Startups
LangChainActiveSeed $25K+; credits need a premier VC
Funded startupsAccelerator startups
Discounted LangSmith seats (Build tier)Best case: $10K credits for 1 year with a premier VC
Build tier: discounted LangSmith seat pricing. Scale tier (premier VC partners only): $10K in LangSmith credits for 1 year, plus events, office hours and promotion.
Eligibility, card, India and how to apply for LangSmith for Startups
- Who qualifies
- First-time LangSmith customer
- Build: less than $10M funding and at least $25K raised in seed funding
- Scale: Series A or earlier (up to $50M) and backed by a premier VC partner
- LangSmith account created with your startup-domain email
- How long
- Scale credits: 1 year (2 years for Y Combinator companies)
- Referral or funding needed
- Build: at least $25K seed raised, under $10M total (any VC). Scale: backed by a LangChain premier VC partner.
- Card needed
- Not stated
- Work or school email
- Yes
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Create a LangSmith account with your startup email and apply for Build or Scale on the startups page.
- Watch out for
- Build tier gives a seat discount, not credits; the discount size is not published.
- Fully bootstrapped startups (no seed money) do not meet the Build tier rule.
- Additional eligibility criteria may apply.
- Status note
- Checked 12 October 2026. Credits are available within a few days of approval.
Official page for LangSmith for Startups (opens in a new tab)Checked 12 October 2026
W&B 50% off Pro for Microsoft for Startups members
CoreWeave (Weights & Biases)Limited time, no end date givenMicrosoft for Startups members only
Bootstrapped startupsFunded startups
50% off W&B Pro for one yearBest case: Advertised as worth $3,000 (real saving varies)
Microsoft for Startups members can get 50% off the W&B (now CoreWeave Forge) Pro plan for one year, advertised as worth $3,000. New customers only, for a limited time.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for W&B 50% off Pro for Microsoft for Startups members
- Who qualifies
- Member of Microsoft for Startups (the page title references Founders Hub; the offer text itself does not restate the membership rule)
- New W&B customers only
- The Pro plan is meant for teams with fewer than 50 employees
- How long
- 12 months
- Referral or funding needed
- Requires Microsoft for Startups membership (offer is listed on W&B’s Microsoft for Startups partner page)
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated on the W&B partner page. Eligibility depends on being accepted into Microsoft for Startups.
- How to apply
- On W&B’s Microsoft for Startups partner page, click ‘Redeem offer’. It opens the CoreWeave Forge startup-program link with the partner coupon applied.
- Watch out for
- This is a 50% discount, not credits. You still pay the other half of the Pro price
- Existing W&B customers are excluded
- The $3,000 value does not match the Pro base price on 12 October 2026: half of $60 a month for 12 months is $360. The real saving depends on your seats and usage
- The page does not give seat or storage limits for the discounted plan (standard Pro: up to 10 seats, 100 GB a month storage)
- Status note
- The W&B partner page says the offer is for a limited time but gives no end date. There is no standalone W&B startup program page any more (wandb.ai/site/startups returns 404). The redeem link redirects to forge.coreweave.com/wandb/startup-program with the coupon applied. Pro now starts at $60 a month with a 30-day free trial. A second check on 13 October 2026 could not reach a live official page for this offer either.
Official page for W&B 50% off Pro for Microsoft for Startups members (opens in a new tab)Checked 12 October 2026
AssemblyAI Startup Program
AssemblyAIActivePre-seed to Series A voice startups
Bootstrapped startupsFunded startupsAccelerator startups
12 months of recurring speech credits (amount unpublished)
12 months of recurring free credits for speech-to-text, streaming and Dictation APIs, plus discounted startup pricing that continues after credits run out. Credit amount not published.
Eligibility, card, India and how to apply for AssemblyAI Startup Program
- Who qualifies
- Pre-seed through Series A; bootstrapped teams welcome
- Product with voice or speech at its core
- In production or launching within six months
- How long
- 12 months of recurring credits; standard pricing phases in afterwards
- Referral or funding needed
- No; Accel, Y Combinator and other partner portfolio companies are fast-tracked
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Fill in the short startup program form on AssemblyAI’s site; credits land in the console within 48 hours of approval.
- Watch out for
- Startup discounts do not apply to LLM Gateway.
- The page says only that the credit is a fixed amount, so ask before you plan around it.
- Status note
- Reviews within one week (48 hours for partner portfolio companies). Separately, the standard free tier needs no card and covers up to 185 hours of pre-recorded and 333 hours of streaming transcription.
Official page for AssemblyAI Startup Program (opens in a new tab)Checked 12 October 2026
Community GPU Grants for Spaces
Hugging FaceActiveCase-by-case grant for Spaces
Open-source maintainersResearchersSolo developersStudents
Free GPU hardware upgrade for a Space demo
Free GPU hardware upgrades for innovative Hugging Face Spaces (demos), granted case by case. Value not published.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for Community GPU Grants for Spaces
- Who qualifies
- An existing Space with an innovative or useful demo
- Apply from the Space’s hardware settings
- How long
- Not stated
- Referral or funding needed
- No
- Card needed
- Not applicable
- Work or school email
- No
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Open your Space’s settings and use the community GPU grant request under the sleep time settings.
- Watch out for
- New compute Spaces (Gradio or Docker) need a paid plan before you can host them.
- Free CPU Basic Spaces sleep after 48 hours without use.
- PRO ($9 a month) includes only $2 of inference credits a month.
- Status note
- Docs checked 12 October 2026. They also say creating a new Space that runs on compute (Gradio or Docker) now requires a paid plan; static Spaces are free. H100 Space hardware was removed in December 2025.
Official page for Community GPU Grants for Spaces (opens in a new tab)Checked 12 October 2026
ElevenLabs Grants (startup program)
ElevenLabsActiveOpen to startups under 25 staff
Bootstrapped startupsFunded startups
12 months free with 33M characters ($5,500+ value)
12 months of free access to the ElevenLabs platform with 33 million characters (over 680 hours of Agents usage, which ElevenLabs values at $5,500+).
Eligibility, card, India and how to apply for ElevenLabs Grants (startup program)
- Who qualifies
- Startup with fewer than 25 employees
- Product with a business or monetization plan; no one-off projects or campaigns
- Valid business email
- No products for users aged 18 or under
- No agencies, consulting firms or SMBs
- One application per company; paying self-serve customers can apply, existing enterprise customers cannot
- How long
- 12 months, or until the 33M characters are used
- Referral or funding needed
- No
- Card needed
- Not stated
- Work or school email
- Yes
- Countries
- Worldwide (ElevenLabs says startups anywhere in the world can apply)
- India
- Yes. ElevenLabs’ official grants announcement (12 August 2025, still live) says startups with fewer than 25 employees anywhere in the world can apply; the current grants page lists no country limit and ElevenLabs runs a dedicated India business page. Checked 13 October 2026.
- How to apply
- Apply on the grants page with your business email; if accepted, the grant plan and credits are added to the account you applied with.
- Watch out for
- After 12 months, when 10,000 credits are left, or when the 33M characters run out, the plan converts to the Free plan unless you choose otherwise.
- Characters cannot be resold or exchanged for cash; doing so ends the grant.
- ElevenLabs will not sign NDAs for applications.
- Status note
- Rolling applications with no deadline; decisions within one week. The same offer appears in Anthropic’s Claude Startup Stack. (Corrected 13 October 2026 after a second check.)
Official page for ElevenLabs Grants (startup program) (opens in a new tab)Checked 13 October 2026
Fireworks for Startups
Fireworks AIActiveVC-backed with $500K+ raised
Funded startupsAccelerator startups
Inference credits for 1 year (amount unpublished)
Credits for on-demand inference, fine-tuning and other Fireworks usage, higher rate limits, access to applied AI engineers, startup events and joint marketing. Credit amount not published.
Eligibility, card, India and how to apply for Fireworks for Startups
- Who qualifies
- Registered, privately held, for-profit company with a working website
- Founded within the last 5 years
- Venture-backed with more than $500K in funding
- Not raised Series D or later
- How long
- Credits expire after 1 year
- Referral or funding needed
- Needs VC backing of more than $500K; no referral code mentioned
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Fill in the form on the Fireworks for Startups page (fireworks.ai/startup-program) and create a Fireworks account; questions to startups@fireworks.ai.
- Watch out for
- Credits cover on-demand usage only, not reserved GPU capacity, SLAs, professional services or premium support.
- After credits run out, billing continues pay-as-you-go if a payment method is on file.
- Status note
- Program page live on 12 October 2026; amounts not disclosed.
Official page for Fireworks for Startups (opens in a new tab)Checked 12 October 2026
Modal for Startups
ModalActivePartner VC backing or $1M+ raised
Funded startupsAccelerator startups
Serverless GPU credits (amount not published)
One-time grant of serverless GPU and compute credits (‘thousands of free GPU credits’), plus technical support and a founder community. Amount depends on traction, funding and VC partner status.
Eligibility, card, India and how to apply for Modal for Startups
- Who qualifies
- New to Modal
- Seed to Series A tier: any VC funding from a partner fund, or more than $1M from any fund
- One grant per tier; no reapplying after receiving credits in a tier
- How long
- Credits last 1 year and cannot be reinstated
- Referral or funding needed
- Yes: funding from one of Modal’s 40+ partner VCs/accelerators (any amount), or more than $1M from any VC. Series B+ custom tier needs over $30M raised or post-Series B, plus a partner VC.
- Card needed
- Yes
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page. Bootstrapped founders can still use Modal’s $30 a month of free compute.
- How to apply
- Apply with the startup credits form on modal.com/startups; email startups@modal.com if there is no reply within 2 weeks.
- Watch out for
- A payment method is required so apps keep running after credits run out.
- Credits cannot pay team or enterprise subscription fees.
- Expired credits are not reinstated.
- Status note
- The page shows no dollar figure. Partners include a16z, Accel, Sequoia, YC, Lightspeed, Antler, General Catalyst and others.
Official page for Modal for Startups (opens in a new tab)Checked 12 October 2026
Nebius for Startups
NebiusChanged: credits only via partner VCsNeeds $5M+ from a Nebius partner VC
Funded startupsAccelerator startups
No credits without a partner VC; founder community onlyBest case: GPU credits (unpublished) via a partner VC
GPU cloud credits, compute discounts and engineering support for AI startups. Credits now go only to startups that raised $5M+ from a Nebius VC partner. Amounts are not published.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for Nebius for Startups
- Who qualifies
- VC- or accelerator-backed, with $5M+ raised from an approved Nebius VC partner
- Building or scaling an AI-enabled product or model
- Working company website, and you apply with a company email
- Incorporated less than 5 years ago
- Not eligible: consulting or outsourced development firms, resellers or distributors, cloud providers, crypto-only businesses, public companies
- How long
- Not stated
- Referral or funding needed
- Yes: at least $5M raised from an approved Nebius VC partner, and you apply through that VC’s platform team
- Card needed
- Not stated
- Work or school email
- Yes
- Countries
- No country limit stated
- India
- Unclear. No country restriction is listed, but credits come only through an approved Nebius VC partner, and the partner list is not public. Most bootstrapped or small-round Indian startups will not qualify. The free founder community is open globally.
- How to apply
- Ask your lead investor’s platform team whether they are a Nebius partner. If so, apply through them via nebius.com/startups. Others can join the free founder community from the same page.
- Watch out for
- Response times are not guaranteed, and applicants may be put on a waitlist
- Startups without a partner VC get no credits, only the founder community
- Third-party claims of $150K plus $10K in credits are not supported by the official page
- Status note
- Directory sites still list ‘up to $150K in GPU credits plus $10K in inference credits’ for NVIDIA Inception members (the earlier ‘AI Lift’ offer). On 12 October 2026 the official page gave no credit amounts, did not mention Inception or AI Lift, and said credits are available only through VC partners for startups that raised $5M+. Applicants may be put on a waitlist.
Official page for Nebius for Startups (opens in a new tab)Checked 12 October 2026
NVIDIA Inception
NVIDIAActiveIncorporated, under 10 years old
Bootstrapped startupsFunded startupsAccelerator startups
Free membership: partner cloud credits, DLI courses
Free program for AI startups: free cloud credits from NVIDIA and partners, preferred pricing on select NVIDIA hardware and software, free self-paced DLI courses and investor introductions.
Eligibility, card, India and how to apply for NVIDIA Inception
- Who qualifies
- At least one developer on staff
- A working website
- Officially incorporated (incorporation date required)
- Less than 10 years old
- Pitch deck with the application
- Revenue and NVIDIA GPU use are not required
- Excludes consultancies and outsourced dev firms, crypto companies, cloud service providers, resellers and public companies
- How long
- Ongoing membership while company details stay up to date
- Referral or funding needed
- No. Any funding stage can apply; no fees, deadlines or equity.
- Card needed
- Not applicable
- Work or school email
- Not stated
- Countries
- No country limit stated (separate contact for China)
- India
- Yes. No country limit is stated. An Indian company must be incorporated before applying.
- How to apply
- Apply through the Inception application portal with company and product details and a pitch deck, then request benefits in the member portal.
- Watch out for
- Preferred pricing is a rebate through suppliers, not a direct sale, and NVIDIA says it cannot guarantee access to specific products.
- Partner cloud credits vary and are not listed publicly.
- The ‘up to $100,000 in DGX Cloud credits’ offer announced in March 2025 was only for selected members and is not on the current program page. NVIDIA’s DGX Cloud page now calls DGX Cloud an internal NVIDIA cloud (checked 13 October 2026). Do not count on it.
- Keep your portal profile updated to stay eligible for benefits.
- Status note
- Checked 13 October 2026. Benefits are requested through the Inception portal. The program page and FAQ list no credit amounts. NVIDIA’s Cloud Partners page still links a March 2025 announcement of up to $100,000 in DGX Cloud credits for selected members, but NVIDIA now describes DGX Cloud as its own internal cloud, so that figure is not treated as a current offer.
Official page for NVIDIA Inception (opens in a new tab)Checked 13 October 2026
OpenAI for Startups
OpenAIActiveNeeds a partner VC referral code
Funded startupsAccelerator startups
No credits without a VC code; Build Hours and eventsBest case: API credits (amount unpublished) with a VC code
Free API credits, rate-limit upgrades and time with OpenAI engineers for startups backed by VCs in OpenAI’s partner network. Others get Build Hours, starter repos and events, but no credit application.
Eligibility, card, India and how to apply for OpenAI for Startups
- Who qualifies
- Backed by a VC in OpenAI’s partner network
- Referral code from that VC
- Product description and planned API use
- Company name, key contacts with a valid business email, and basic funding details
- OpenAI organization ID
- How long
- Not stated; after credits run out the account moves to standard billing
- Referral or funding needed
- Yes. A unique referral code from a partner VC (examples named: Thrive Capital, Sequoia, a16z, Kleiner Perkins, Conviction Partners).
- Card needed
- Not stated
- Work or school email
- Yes
- Countries
- OpenAI API supported countries
- India
- Yes. India is on OpenAI’s API supported-countries list. The catch is the VC route: credits need a referral code from a partner VC, and the VCs named are US firms. Ask your investor whether it is in OpenAI’s network.
- How to apply
- Ask your VC whether it is an OpenAI partner and for its referral code, then apply with the code, your business email and your OpenAI org ID.
- Watch out for
- No credits for bootstrapped or non-partner-backed startups.
- When credits run out, usage switches automatically to standard billing.
- Zero Data Retention is available on request on paid plans, not by default.
- Status note
- Checked 12 October 2026. No self-serve credit application; OpenAI does not publish credit amounts. OpenAI also hands out credits at some events and hackathons (it asks for your org ID there).
Official page for OpenAI for Startups (opens in a new tab)Checked 12 October 2026
Sarvam Startup Program and free API sign-up credits
Sarvam AIActiveOpen: no referral or funding rule
Solo developersBootstrapped startupsFunded startups
Rs 100 sign-up credit; program credits set case by case
Indian-language AI APIs (speech, translation, text). New API accounts get Rs 100 in free credits. The Startup Program gives 6 to 12 months of API credits sized to your scale, plus direct engineering support.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for Sarvam Startup Program and free API sign-up credits
- Who qualifies
- No formal criteria published; the form asks for product stage, funding stage, industry, team size and location
- Existing Sarvam API users can apply for more credits
- How long
- Program credits for 6 to 12 months. Sign-up and paid credits do not expire
- Referral or funding needed
- No requirement stated
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Yes. Bengaluru-based provider. Pricing is in INR (Rs 1 = 1 credit).
- How to apply
- Sign up on Sarvam’s dashboard for the free credits. Apply for the program through the form at sarvam.ai/startup-program.
- Watch out for
- You must show ‘Powered by Sarvam’ in your product as a program condition.
- Unused program credits do not roll over after the term; then standard pricing applies.
- The free sign-up credit is small (Rs 100).
- Program credit amounts are not published.
- Status note
- Pages live as of 12 October 2026. Sarvam’s docs state Rs 100 of free credits on sign-up. The program credit amounts are set case by case and not published.
Official page for Sarvam Startup Program and free API sign-up credits (opens in a new tab)Checked 12 October 2026
Weaviate startup deal
WeaviateActiveForm asks for your backing VC/program
Accelerator startupsFunded startups
Discounted Weaviate Cloud pricing (size unpublished)
Discounted Weaviate Cloud pricing for startups; the redemption form asks which program or VC backs you (examples: Y Combinator, Techstars, partner programs). Discount size not published.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for Weaviate startup deal
- Who qualifies
- Startup with a Weaviate Console account and completed organization settings
- Name of the backing program or VC, and proof if requested
- How long
- Not stated
- Referral or funding needed
- Unclear. The form asks for the backing program or VC, but the page does not say backing is required.
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Log in to Weaviate Console, complete organization settings and submit the redemption form with the same email; Weaviate validates and applies the deal.
- Watch out for
- The discount amount and duration are not stated on the official page.
- Use the same email as your Console account, or explain the difference in the form.
- Status note
- Redemption page live on 12 October 2026. Third-party sites quote large credit figures that are not on Weaviate’s pages.
Official page for Weaviate startup deal (opens in a new tab)Checked 12 October 2026
AWS Generative AI Accelerator
Amazon Web ServicesNot fully confirmedCompetitive cohort; 2026 run unconfirmed
Funded startupsBootstrapped startupsAccelerator startups
Up to $1M in AWS credits if selected for the cohort
A selective cohort program for generative AI startups, offering up to $1M in AWS credits each, plus mentorship and a closing week at AWS re:Invent. The 2025 cohort took 40 startups for 8 weeks.
Some details are uncertain or conflict between official pages. Check the official page before you apply.
Eligibility, card, India and how to apply for AWS Generative AI Accelerator
- Who qualifies
- Startup building generative AI, such as fine-tuning open models, pre-training foundation models, or services in the gen AI stack
- Selection based on idea, technical readiness and interview (per researcher; not re-confirmed on the 2025 article)
- The 2025 cohort spanned Asia Pacific, North America, Latin America, Europe and the Middle East
- How long
- 8-week cohort (2025: started 14 October in Seattle, ended at re:Invent in December); credit validity not stated
- Referral or funding needed
- No referral stated; competitive selection (40 picked from thousands of applicants in 2025)
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- Global in 2025 (APAC, North America, LATAM, Europe, Middle East); no country list published
- India
- Unclear. The 2025 cohort was global and included Asia Pacific, but no country list was found. No 2026 call was confirmed.
- How to apply
- Watch the AWS Startups programs page for a new call; no application form was live on 12 October 2026.
- Watch out for
- ‘Up to $1M’ is a ceiling, and the amount per startup is not published.
- The 2025 cohort met in person in Seattle and at re:Invent in Las Vegas.
- AWS Activate is the regular route for credits outside a cohort.
- A China-only ‘AWS Startup Accelerator Cohort 4’ (up to $100K Bedrock credits) is a different program.
- Status note
- On 12 October 2026 the old program URL (aws.amazon.com/startups/generative-ai/accelerator/) redirects to AWS’s 7 October 2025 article on the 2025 cohort, which confirms 40 startups, 8 weeks and up to $1M in credits. A second old URL (aws.amazon.com/startups/accelerators/generative-ai) redirects to startups.aws.com/programs/generative-ai, which returns 404. The AWS Startups programs page lists only the ‘AWS Inference Fellowship’ under Accelerators. No official page confirms a 2026 cohort; a third-party site claims a mid-2026 window, unverified. Equity is not mentioned.
Official page for AWS Generative AI Accelerator (opens in a new tab)Checked 12 October 2026
Llama Startup Program
MetaNot fully confirmedUS-only cohort; page now offline
Bootstrapped startupsFunded startups
Reported up to $6,000 a month in cloud-cost reimbursementBest case: Up to $36,000 over six months (reported)
Launched in May 2025 for US startups building on Llama: support from Meta’s Llama team and up to $6,000 a month for up to six months in cloud-cost reimbursement. The official page no longer loads.
Low confidence: we could not confirm the key details on an official page. Check the official page before you rely on this entry.
Eligibility, card, India and how to apply for Llama Startup Program
- Who qualifies
- Reported for the first round: US-based, incorporated, under $10M raised, at least one developer, building generative AI apps; deadline 30 May 2025 (unconfirmed)
- How long
- Reported: up to 6 months (unconfirmed)
- Referral or funding needed
- Reported: under $10M raised; no referral (unconfirmed)
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- Reported first round: US only (unconfirmed)
- India
- Unclear. The 2025 round was reported as US-only, and no later round could be confirmed, so treat it as not open to Indian companies unless Meta announces otherwise.
- How to apply
- No live application page found.
- Watch out for
- Funding was reported as reimbursement of cloud costs, not upfront cash.
- Status note
- On 12 October 2026 llama.com/llama-startup-program redirected to developer.meta.com, which showed an error page. A first cohort was announced in July 2025 (press); no 2026 cohort found. A second check on 13 October 2026 could not reach a live official page for this offer either.
Official page for Llama Startup Program (opens in a new tab)Checked 12 October 2026
Groq for Startups
GroqNot fully confirmedSelected startups; program page gone
Funded startupsAccelerator startups
Reported $10,000 inference credits (unconfirmed)
Previously advertised $10,000 in GroqCloud inference credits for selected startups. The official program page now redirects to Groq’s home page.
Low confidence: we could not confirm the key details on an official page. Check the official page before you rely on this entry.
Eligibility, card, India and how to apply for Groq for Startups
- Who qualifies
- Unconfirmed: hand-selected startups and VC partners
- How long
- Reported as 90 days after award (unconfirmed)
- Referral or funding needed
- Unclear; Groq selected startups at its discretion
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- Unknown
- India
- Unclear. Program status unknown; nothing official to check against.
- How to apply
- No live application page found; ask Groq sales or your VC whether the program still runs.
- Watch out for
- Do not rely on third-party listings that still show this offer as open.
- Status note
- On 12 October 2026 groq.com/groq-for-startups redirected to groq.com. The $10,000 amount and 90-day expiry come only from third-party listings; one 2026 roundup lists the program as ‘appears withdrawn’. A second check on 13 October 2026 could not reach a live official page for this offer either.
Official page for Groq for Startups (opens in a new tab)Checked 12 October 2026
Perplexity for Startups
PerplexityNot fully confirmedNeeds a Startup Partner; page gone
Accelerator startupsFunded startups
Reported $5,000 Sonar API credits + Enterprise Pro
Launched in April 2025 with $5,000 in Sonar API credits and six months of Enterprise Pro for up to 50 seats. The official page now redirects to Perplexity’s hub page.
Low confidence: we could not confirm the key details on an official page. Check the official page before you rely on this entry.
Eligibility, card, India and how to apply for Perplexity for Startups
- Who qualifies
- Reported at launch: under 5 years old, under $20M equity raised, linked to an approved Startup Partner, not an existing Enterprise Pro customer (unconfirmed)
- How long
- Reported: 6 months of Enterprise Pro (unconfirmed)
- Referral or funding needed
- Reported: must select an approved Perplexity Startup Partner when applying (unconfirmed)
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- Unknown
- India
- Unclear. Program status unknown; some partner programs (including India-focused perk sites) still list it, but nothing official was available to check.
- How to apply
- No live application page found; ask your accelerator or VC whether they still offer it.
- Watch out for
- Directories still list this offer; confirm with Perplexity before relying on it.
- Status note
- On 12 October 2026 perplexity.ai/startups redirected to perplexity.ai/hub. All amounts and rules come from April 2025 press coverage and partner listings, not an official page. A second check on 13 October 2026 could not reach a live official page for this offer either.
Official page for Perplexity for Startups (opens in a new tab)Checked 12 October 2026
Ended and paused programs (1)
Vapi for Startups (startup grant)
VapiPaused$250K+ raised; applications on hold
Funded startupsAccelerator startups
7,500 voice minutes a month for 12 months (paused)
When open: 7,500 free voice-agent minutes a month for 12 months (90,000 minutes), a private Slack channel with Vapi engineers and early feature access. Applications are on hold.
Eligibility, card, India and how to apply for Vapi for Startups (startup grant)
- Who qualifies
- Building voice AI products or features
- Pre-seed to Series A with $250K+ raised
- Launching to customers within 6 months
- Agencies and consulting firms excluded
- How long
- 12 months
- Referral or funding needed
- Needs at least $250K raised (pre-seed to Series A); no referral needed
- Card needed
- Not stated
- Work or school email
- Not stated
- Countries
- No country limit stated
- India
- Unclear. No country limit is stated; we did not confirm India sign-up on an official page.
- How to apply
- Applications are on hold; watch vapi.ai/startups for reopening.
- Watch out for
- Regular users get 100 free minutes; the 7,500 a month applied only to accepted startups.
- Above 90,000 minutes, or after 12 months, sales contacts you to move to a paid plan.
- Vapi does not give rejection reasons, and meeting the criteria does not guarantee entry.
- Status note
- On 12 October 2026 vapi.ai/startups shows ‘Applications on hold’ on both apply buttons, with no reopening date. The program terms are still listed.
Official page for Vapi for Startups (startup grant) (opens in a new tab)Checked 12 October 2026
Free GPU credits and free GPU notebooks
If you need a GPU for learning or experiments, start with a free notebook. If you run a funded startup, apply for credits. Most “free GPU credit” offers we found on 12 and 13 October 2026 are startup programs with a funding rule.
Free GPU notebooks
- Kaggle: 30 GPU hours a week (sometimes more, depending on demand) and 20 TPU hours (TPUs are Google’s own AI chips). Sessions run up to 12 hours, or 9 for TPU, on one P100 or two T4s. Interactive sessions end after 20 minutes idle, so use “Save & Run All” for long jobs.
- Google Colab: free GPUs when available, but not guaranteed; at busy times you may get only a CPU. Sessions run up to 12 hours. Google publishes no fixed free GPU-hour figure.
GPU credits for startups
Grouped by what you need to get in. Full terms are in each entry.
- No funding needed: E2E Networks AI Explorer ($5,000 for 12 months; 10 or fewer staff, founded in the last 5 years, from any country) and Yotta Launch ($10,000 paid in tranches; the first is $1,000).
- A working product, pre-Series A: Runpod Starter ($1,000; venture backing strongly preferred). Its $25,000 Growth bonus needs a $50,000 upfront commitment.
- Selection, no referral: Together AI (up to $15,000 if you raised up to $5M; more for larger rounds).
- Seed to Series A: Baseten (up to $25,000 plus $2,500 for Model APIs, expiring after 6 months).
- A partner VC or a large round: Modal (a partner VC or more than $1M raised; amount not published) and Nebius (medium confidence; $5M+ from a partner VC, whatever older directory listings say).
If you qualify for none of these, you can still use Modal’s $30 a month of free compute.
Research and demo grants
- Lambda Research Grant: up to $5,000 in Lambda cloud credits for qualifying researchers. More research credits, including OpenAI’s and Anthropic’s, are on our credits page for researchers, nonprofits and open source.
- Community GPU Grants for Spaces (medium confidence): free GPU upgrades for innovative Spaces demos, granted case by case. On 12 and 13 October 2026, Hugging Face’s docs said new Gradio or Docker Spaces need a paid plan before you can host them.
India. The IndiaAI compute portal offers subsidised GPUs, not free ones: the official calculator shows a maximum subsidy of up to 40%, subject to approval. See IndiaAI compute and Indian GPU credits.
Free AI coding agents in 2026
AI coding agents are tools that read your code and write or change it for you, using AI models behind the scenes. Four had a $0 plan on 12 and 13 October 2026:
- GitHub Copilot Free (medium confidence): 2,000 code completions a month plus a small allowance of GitHub AI Credits. GitHub’s docs have not stated the size of that allowance since usage-based billing began on 1 June 2026, though its plans page FAQ still says 50 chat requests a month. No card needed.
- Google Antigravity Individual: $0 for the agent IDE and CLI, with a weekly Gemini quota Google does not publish. It replaced the free Gemini CLI and Code Assist tier on 18 June 2026. You must be 18 or over; India is on the supported list.
- Jules: 15 tasks per rolling 24 hours, 3 at a time, on Gemini 2.5 Pro. It works on your GitHub repos.
- Kiro Free: 50 credits a month, on open-weight models (models anyone can download and run) and Claude Sonnet 4.5. Agent tasks use different amounts per prompt, so 50 credits go fast.
Students get more: Copilot Student, Kiro Students and the Codex credits are on our free credits page for students. Copilot Pro for teachers and maintainers, and Cursor’s student credits, are compared on our free AI plans page. If you prefer a terminal agent, our Claude Code tutorial walks through setup.
Free AI APIs that ended: do not follow old tutorials
Many tutorials and videos still point to these. They no longer work as described:
- GitHub Models: retired for everyone on 30 July 2026, including its playground and inference API. GitHub points users to Azure AI Foundry and GitHub Copilot.
- Free Gemini CLI and Code Assist for individuals: stopped serving requests on 18 June 2026. The old “1,000 requests a day” figure is gone; use Antigravity, or Gemini CLI with a paid Gemini API key.
- Amazon Q Developer Free: closed to new sign-ups on 15 May 2026, and its pricing page still shows the free tier. Kiro replaces it.
- Cerebras: replaced its free tier with a $5, 30-day trial. Its pages give no date for the change.
We could not confirm these. Older lists still carry them, but on 12 and 13 October 2026 we found no official page that still describes the offer. Their entries show the amounts older lists report, marked as unconfirmed. Do not plan around them:
- Grok API free credits (data-sharing credits): xAI’s billing docs mention only prepaid credits and promo codes.
- Groq for Startups: the official page redirects to Groq’s home page.
- Llama Startup Program: the official page no longer loads.
- Perplexity for Startups: the official page redirects to Perplexity’s hub page.
What free AI tiers do with your data
The table at the top shows what each free tier said about your prompts on 12 and 13 October 2026. They fall into three groups:
- Used to train models or read by people: the Gemini API free tier (prompts and outputs improve Google products and may be read by human reviewers) and OpenAI’s data-sharing tokens (shared inputs and outputs can train OpenAI’s models). xAI’s reported free credits, which we could not confirm, also needed data sharing.
- A setting you control: Mistral’s “Anonymous improvement data” toggle under Privacy in the Admin panel, and OpenRouter’s model-training toggle, which decides whether providers that log prompts are used. Check both before you send anything sensitive.
- Not stated: the GroqCloud, Cohere, NVIDIA, Cloudflare, Cerebras and Hugging Face pages we checked say nothing either way. Read each provider’s privacy policy before you send real data.
For finance and accounting teams: never send client, payroll, tax or bank data to a free tier. That includes salary sheets, GST or VAT returns, bank statements and anything with a customer’s name on it. Test with made-up data on a free tier, then move real work to a paid tier with the data terms your clients and regulators expect.
Professional help
Got credits? Put them to work on your finance workflows
We build automation for finance and accounting teams: invoice capture, reconciliations and reports that update themselves. Tell us what slows your books down and we will reply with a plan and a quote.
Invoice and document capture with AI, checked by a person
Bank, GST and vendor reconciliations that run on a schedule
MIS reports your team can trust, built on your own data
No spam. We use your details only to reply to this request.
Free download
Startup credits tracker for Excel
Log each credit grant with its expiry date and how much you use each month, and see which credits will run out or expire first. The workbook also lists the programs we track, with a link to each official page.
Download the tracker (.xlsx)Free Excel file. No email or sign-up needed. Programs checked 12 and 13 October 2026.
Free AI API credits: FAQ
What are the Gemini API free tier rate limits?
Google shows the exact free-tier limits only inside AI Studio, for your own project. Daily request quotas reset at midnight Pacific time. On 12 and 13 October 2026 the free tier covered Gemini 3.x Flash and Flash-Lite, 2.5 Pro and Flash, Gemma 4 and embeddings.
Is the Groq API free?
Yes, with limits. GroqCloud’s Free plan needs no card. On 12 and 13 October 2026 it allowed 30 requests a minute, 1,000 requests a day, 8K tokens a minute and 200K tokens a day on gpt-oss-120b. Other models have their own limits, shown on the Limits page for your organisation. Higher limits, Batch and Flex need the paid Developer plan, which needs a credit card, a US bank account or a SEPA debit account.
Can I use the Google Cloud $300 free trial for the Gemini API?
No. Google’s Gemini API pages said on 12 and 13 October 2026 that the Google Cloud $300 trial credit cannot pay for Gemini API use in AI Studio. Use the free tier, or move to paid use, where new users start on Prepay with a $5 minimum top-up. The $300 trial still works on other Google Cloud products; see our free cloud credits page.
Can I use a free AI API tier in production?
Not for anything customers rely on. Cohere’s Trial key may not be used for production or commercial use, NVIDIA’s free endpoints are for development and testing, and Mistral’s Free mode is meant for evaluation and prototyping. In the EEA, Switzerland and the UK, Google allows only paid Gemini services in apps for users there. Most other free tiers do not say either way, and their limits can change without notice, so use a paid tier.
Are free API key websites safe to use?
No. A key you did not create in the provider’s own console is someone else’s key or a proxy that can read everything you send, and it can stop working at any moment. Create keys only in official consoles such as Google AI Studio, GroqCloud, OpenRouter, Mistral Studio and the OpenAI and Claude consoles.
Where can I get free GPU credits?
Without a company, Kaggle gives 30 GPU hours a week and Colab gives free GPUs when available. Startups can apply for Runpod’s $1,000 Starter tier, up to $15,000 from Together AI on its Build tier or $5,000 from E2E Networks; researchers can apply for Lambda’s grant of up to $5,000. All checked 12 and 13 October 2026.
Is GitHub Models still free?
No. GitHub retired GitHub Models for everyone on 30 July 2026, including the playground and inference API. GitHub points users to Azure AI Foundry and GitHub Copilot instead.
Related guides
- AI leaderboard: compare models before you spend free credits
- Monthly API credits on Claude Max and Team
- Cloudflare for Startups: tiers, Workers AI credits and how to apply
- Claude Code tutorial
- Codex for finance teams
- Free ChatGPT and Gemini app plans
- Cloud and AI credits for startups
- Free cloud credits: Google Cloud, Azure and AWS trials
- Copilot Student and Codex credits for students
- Every free credit program we track, A to Z
Please note: AI free tiers and credit programs change their limits often, sometimes without notice, and there can be a gap between a change and our update. Every limit on this page is as shown on the official pages on 12 and 13 October 2026. Check the provider’s own limits page or console before you rely on a figure. Credit amounts are the providers’ advertised values in US dollars unless stated, not cash, and nothing here is tax or legal advice.
Sources
- Gemini API pricing: free and paid-only models (Gemini 4 Argon paid-only, checked 15 October 2026), free-tier data use, 1 January 2027 price changes (opens in a new tab)
- Gemini API rate limits: free tier requirements, quota reset, limits shown in AI Studio (opens in a new tab)
- Gemini API additional terms: age, unpaid services data use, EEA, Switzerland and UK rules (opens in a new tab)
- Gemini API available regions (opens in a new tab)
- Gemini API billing: Prepay and Postpay, which Cloud credits apply, $300 Welcome credit (opens in a new tab)
- Gemini API Flex inference: half price, slower, best-effort (checked 15 October 2026) (opens in a new tab)
- Google Developer Program: monthly Google Cloud credits with Google AI Pro and Ultra (opens in a new tab)
- GroqCloud rate limits: Free and Developer plans (opens in a new tab)
- GroqCloud billing FAQ (opens in a new tab)
- OpenRouter limits: free model variants (opens in a new tab)
- OpenRouter FAQ: fees, credit expiry, logging and training toggle (opens in a new tab)
- Mistral Studio: Free mode and API keys (opens in a new tab)
- Mistral help: opting out of training on API data (opens in a new tab)
- Cohere pricing: Trial and production keys (opens in a new tab)
- NVIDIA API Catalog: models and free endpoints (opens in a new tab)
- Cloudflare Workers AI pricing: free daily Neurons (opens in a new tab)
- Cerebras rate limits: Free Trial and no renewing free tier (opens in a new tab)
- Cerebras pricing: $5 trial credit (opens in a new tab)
- Hugging Face Inference Providers pricing: monthly credits by account type (opens in a new tab)
- OpenAI help: complimentary daily tokens for sharing data (opens in a new tab)
- OpenAI for Startups (opens in a new tab)
- Claude help: monthly API credits for Max and Team plans (opens in a new tab)
- Claude for Startups program page and FAQ (checked 15 October 2026: cohort full, no new applications) (opens in a new tab)
- AWS Activate terms: third-party models on Amazon Bedrock (opens in a new tab)
- Google for Startups Cloud Program FAQ (opens in a new tab)
- DigitalOcean Startups: qualifying services since 13 May 2026 (opens in a new tab)
- Vercel for Startups credits: 50% AI Gateway cap (opens in a new tab)
- Cloudflare for Startups: Workers AI caps and AI Gateway (opens in a new tab)
- Kaggle notebooks: GPU and TPU quotas and sessions (opens in a new tab)
- Google Colab FAQ (opens in a new tab)
- GitHub docs: GitHub Models retirement (opens in a new tab)
- Google: Gemini Code Assist for individuals deprecation (opens in a new tab)
- AWS: Amazon Q Developer end of support (opens in a new tab)
- GitHub Copilot plans (opens in a new tab)
- Google Antigravity plans: free weekly quota (opens in a new tab)
- Jules usage limits (opens in a new tab)
- Kiro pricing: Free plan credits and models (opens in a new tab)