Name That LLM Kaggle Competition 2026: How to enter for a share of $50,000

The Name That LLM Kaggle competition splits $50,000 equally among the ten teams that best identify which AI model wrote a short passage of text. Each of the top 10 places on the private leaderboard wins $5,000, and the only thing that counts is balanced accuracy across 18 frontier models from OpenAI, Google, Anthropic, Qwen, ZAI and Moonshot AI. You must accept the rules before the Entry Deadline of Wednesday 6 January 2027, 11:59 PM UTC (Coordinated Universal Time), and final submissions close on Wednesday 13 January 2027 at 11:59 PM UTC. The dates table below shows both moments in your own local time.

It is a Kaggle code competition, so you do not upload answers. You write a Kaggle Notebook (a free, browser-based coding workspace on Kaggle’s own computers), Kaggle runs it on about 4,000 hidden passages with the internet switched off, and your score appears on a leaderboard. A leaderboard is Kaggle’s ranking table of competitors’ scores. The public leaderboard ranks you on part of the hidden data during the contest, and the private leaderboard ranks you on the rest and decides the prizes. It is free to enter, it is open to residents of the United States and most other countries (a short list of regions is excluded), and you can work alone or in a team of up to five.

This guide covers every date in UTC and in your local time, the exact file Kaggle expects, how the score is calculated, the rules that trip up first-timers, and who owns the code if you win. Every fact comes from the official competition overview, the official rules and Kaggle’s own documentation, as checked on 17 October 2026.

Hackathon guide · Kaggle, sponsored by Google LLC

Name That LLM Kaggle Competition

Open

Time left to joinWed 6 Jan 2027, 23:59 UTC

Last moment to join: Wed 6 Jan 2027, 23:59 UTC

Prize pool

$50,000

Ten prizes of $5,000 each, places 1 to 10

Deadline (UTC)

Wed 13 Jan 2027, 23:59 UTC

January 13, 2027: Final Submission Deadline, 11:59 PM UTC. Accept the rules by January 6, 2027, 11:59 PM UTC.

Who can enter

Worldwide (6 regions excluded), 18 or local age of majority, whichever is older

Excludes Crimea, the so-called DNR and LNR, Cuba, Iran, North Korea and anyone under US sanctions or export controls. Under-18s need consent forms, one Kaggle account each

Format

Online Kaggle code competition

Solo or teams of up to 5, scored automatically

Open the official pageOfficial rulesAll open hackathons

Name That LLM Kaggle Competition: the short answer

What is Name That LLM Kaggle Competition?

Name That LLM: Identify the Model Behind the Text is a Featured Code Competition hosted on Kaggle, an online platform where people compete on data science problems and share datasets and code. The Competition Sponsor named in the official rules is Google LLC of Mountain View, California. A code competition is one where you submit the program that makes the predictions, not the predictions themselves: Kaggle re-runs your code on test data you never see and scores whatever it produces.

The task is authorship attribution, which means working out who wrote a piece of text. Here the authors are 18 frontier large language models (LLMs), the AI systems behind modern chatbots. They come from six labs: OpenAI (gpt-5.5, gpt-5.6-luna, gpt-5.6-sol, gpt-5.6-terra and gpt-6-astra), Google (gemini-3.1-pro-preview, gemini-3.6-flash, gemini-3.7-flash and gemini-3.8-flash), Anthropic (claude-fable-5, claude-opus-4-6, claude-opus-4-8, claude-opus-5 and claude-sonnet-5), Qwen (qwen3.8-max), ZAI (glm-5.3 and glm-5.3-flash) and Moonshot AI (kimi-k3). Your code reads each unlabeled passage, between 40 and 250 words long, and outputs exactly one of those 18 names.

The passages answer prompts on philosophy, physics, mathematics, biology, computer science and creative prose. Kaggle describes them as authentic LLM completions with only light formatting cleanup, such as removing a bold wrapper around a whole passage, and it warns that character counts and topic keywords will not work as shortcuts. The training file holds 2,699 labelled passages written for 150 prompts, and the hidden test set holds about 4,000 passages written for prompts that never appear in training. Kaggle says the aim is to sharpen methods for AI provenance, the job of tracing where a piece of content came from, and for benchmark auditing.

There is one track and no sub-tracks. The competition opened on 14 October 2026 and runs to 13 January 2027. As of 17 October 2026 the overview showed 4,989 entrants, 1,236 participants, 1,199 teams and 4,873 submissions, and the top public leaderboard score was 0.593 balanced accuracy (the average, across all 18 models, of the share of each model’s passages labelled correctly). For comparison, by our own arithmetic, random guessing across 18 equally weighted models would score about 0.056 (1 divided by 18), so the early leaders were well ahead of chance and still a long way from perfect.

Key dates and deadlines in UTC and your local time

Kaggle states every deadline in UTC (Coordinated Universal Time), the global reference clock, which never changes for daylight saving. The overview says all deadlines are at 11:59 PM UTC on the day shown unless otherwise noted. That means the official cutoff is one minute before midnight UTC, not one minute before midnight where you live.

Depending on your time zone, that moment can fall on a different calendar day. West of UTC, across the Americas, an 11:59 PM UTC deadline lands in the afternoon or evening of the same day: the final deadline is 3:59 pm PST (UTC-8) or 6:59 pm EST (UTC-5) on Wednesday 13 January 2027 in the United States. East of UTC it lands on the next day: 12:59 am CET (UTC+1) in Central Europe or 8:59 am JST (UTC+9) in Japan, both on Thursday 14 January. The table gives each date as written, in UTC with the weekday, and in your own local time, worked out automatically for your device.

One inconsistency is worth knowing. The written Timeline on the overview, which Rules section 2.3 points to, gives 13 January 2027 as the Final Submission Deadline. The countdown marker labelled Close on the same page is set one day later, at 11:59 PM UTC on Thursday 14 January 2027. Plan for 13 January. If the extra day turns out to be real, treat it as a safety buffer, not a target.

The Entry Deadline line says you must accept the rules before 6 January 2027, while deadlines in general fall at 11:59 PM UTC on the day shown. Do not test that wording. Join the competition and accept the rules well before 6 January, even if you plan to start building later.

The written Timeline gives no start time, but the Start marker on the overview shows the competition opened at 13:15 UTC on Wednesday 14 October 2026. The opt-out date in the table is our own calculation from the rule that a potential winner who does not want a prize must say so within one week after the Final Submission Deadline, so it moves by a day if the later close date applies.

Prizes

The prize pool is $50,000, paid as ten equal prizes of $5,000 each. Places 1 to 10 on the final private leaderboard each win the same amount, so tenth place pays exactly as much as first. Every figure is before tax: prizes are paid net of any taxes the sponsor is required by law to withhold, and winners pay all taxes on their prizes.

The rules write amounts with a $ sign and do not spell out a currency code. The sponsor, Google LLC, is based in the US. The rules also do not state the payment method, so if you live outside the US, how the prize reaches you and in what currency is something to confirm with Kaggle if you win.

Rank is decided only by balanced accuracy on the private leaderboard. A team prize is split evenly among eligible members, unless the whole team unanimously agrees a different split and tells Kaggle before the prize is paid. Prizes cannot be transferred or assigned to anyone else, and they are paid net of any taxes the sponsor is required by law to withhold.

The competition also awards Kaggle Points and Medals, the platform’s own recognition for competitors. Employees, interns, contractors, officers and directors of Google LLC, Kaggle Inc. and their parents, subsidiaries and affiliates can take part but cannot win a prize. If a potential winner is disqualified, the next highest score is chosen instead.

Who can enter, and from which countries

The competition runs under the Kaggle Foundational Rules plus a set of competition-specific rules, and the Foundational Rules win wherever the two conflict. You need one registered Kaggle.com account, and only one: entering or submitting from more than one account gets you disqualified. You must be 18 or the age of majority where you live, whichever is older, unless the sponsor agrees otherwise and a parent or guardian has given consent.

A pinned Kaggle Staff post on the competition forum says under-18s can take part by submitting two parent or guardian consent forms, one for using Kaggle and one for this competition. Children under 13 are not allowed on Kaggle at all. Students and working professionals can both enter, and there is no student-only or professional-only rule. If you enter on behalf of your employer, you confirm that your employer knows and consents.

The official rules open the competition to residents of the United States and worldwide, with these exceptions: residents of Crimea, the so-called Donetsk People’s Republic (DNR), the so-called Luhansk People’s Republic (LNR), Cuba, Iran and North Korea, and anyone subject to US export controls or sanctions. Local laws can still apply, and a winner who lives in a country where prizes cannot be awarded cannot receive one.

Teams and solo entries

Everyone competes as a team on Kaggle, even a team of one. A team can have at most five members, each person can join or form only one team, and every member needs their own Kaggle account. You register individually first, then confirm your membership by responding to the team notification sent to your Kaggle account.

The team leader handles mergers, and the last day to join or merge teams is Wednesday 6 January 2027, 11:59 PM UTC. Mergers also carry a submission cap. The competition-specific rule measures it as of the Team Merger Deadline, while the Foundational Rules, which take priority in a conflict, measure it on the date of the merger request. In practice, the merging teams’ combined submissions cannot exceed what one team could have made by the day you ask to merge, which is 5 a day multiplied by the number of days the competition has run, and the merger must be complete by the merger deadline.

The limit of 5 submissions a day applies to the whole team, not to each member. Sharing code or data privately with anyone outside your team is not allowed. If you share code publicly, the rules say it must go on the competition’s Kaggle forum or in public Kaggle notebooks. Kaggle Staff say the official Kaggle Discord, a group chat server, counts as public, but staff do not monitor it, so keep code and important questions on the Kaggle forum.

Prize money is split evenly among eligible members unless the team unanimously agrees a different split and tells Kaggle before prizes are issued. Our advice: agree the split inside your team early, before anyone is on the leaderboard. If your teammates live in different countries, remember that each winner’s paperwork and tax position depends on where they live.

How to enter Name That LLM Kaggle Competition, step by step

Here is the full path from a new account to a scored final submission, following Kaggle’s own pages. A Kaggle Notebook is a free, browser-based coding environment where you write and run code on Kaggle’s computers instead of your own.

Do steps 1 and 2 straight away, even if you plan to build later. The entry cutoff on 6 January 2027 comes a full week before the final deadline.

What you must submit

Because this is a code competition, what you submit is a notebook, not a file of answers. Kaggle’s competitions documentation says all submissions must be made from inside a Kaggle Notebook, and predictions cannot be uploaded directly. Your notebook has to meet every requirement below or the submission fails.

You do not submit a video, demo link, code repository or written description when you enter. The rules do not state a required programming language. Kaggle Notebooks support Python and R, and the official example uses Python with pandas, a popular library for working with tables of data.

If you finish in the top 10, more is required. Winners must deliver training code, inference code (the code that makes predictions with a trained model), documentation and a description of the computing environment needed to run it. The sponsor may also ask for a detailed, reproducible write-up of how the winning submission was generated, with a link to a code repository, and may ask you to join a recorded or panel call. Clean, commented code from day one makes all of that easy.

How entries are judged

There is no judging panel. Scoring is automatic and uses a single metric, balanced accuracy: the average, across all 18 models, of the share of each model’s passages you labelled correctly. That per-model share is called recall. Because every model counts equally, getting a rarely seen model right matters exactly as much as getting a common one right.

The test set does not hold the same number of passages for each model, and the mix of models differs between the public and private leaderboards. A classifier that leans toward the most common labels can look fine on plain accuracy and still score poorly here.

Tools, APIs and what they cost

The only required tool is a Kaggle Notebook, and it is free. Per Kaggle’s notebook documentation, each session gets 4 CPU cores and 30 GB of RAM, or a GPU option with one NVIDIA Tesla P100 or two T4s and 29 GB of RAM. A GPU (graphics processing unit) is a chip that speeds up neural network training and inference. Kaggle’s GPU usage guide says the free GPU quota resets weekly and is 30 hours, or sometimes higher depending on demand and resources. Code competition scoring runs always use Intel Skylake CPUs, and for scoring your notebook must have internet off and finish within 9 hours.

No specific model, API or paid service is required. An API, or application programming interface, is a way for code to call another company’s service over the internet. Since scoring runs have the internet switched off, your notebook cannot call a hosted model at scoring time anyway. Freely available pre-trained models and public data are allowed. Paid data or tools must meet the rules’ Reasonableness Standard, which treats a small subscription (the rules use Gemini Advanced as the example) as acceptable and a proprietary dataset that costs more than a prize as not acceptable.

No credits or free API tiers are offered for this competition. Colab Pro and Pro+ subscribers can link their accounts to Kaggle for 15 or 30 extra GPU hours a week depending on the plan, but that is a general Kaggle offer, not part of this competition. As of 17 October 2026, participants had asked on the forum whether extra training data generated with the candidate models’ paid APIs, chat subscriptions or the free Kaggle Benchmarks AI quota is allowed, and no host reply was visible. The rules only say external data must be publicly available and equally accessible at no cost, or meet the Reasonableness Standard, so treat generated data as an open question until the host answers on the forum.

Rules that trip people up

These are the rules and quirks most likely to cost you a submission, a ranking or a prize. Each one comes from the official rules, the competition pages or Kaggle’s documentation.

Ownership, prize payment and tax

You warrant that your submission is your own original work and that you own it. If you win, the competition-specific terms list the winner licence type as CC-BY 4.0, a Creative Commons licence that lets others reuse work as long as they give credit. Section 2.5 of the rules also says winners license the winning submission and the source code used to generate it under an Open Source Initiative (OSI) approved licence that does not limit commercial use. The rules do not explain how those two licence statements fit together. Commercially available software, and data or pre-trained models with incompatible licences, are carved out.

Any open-source code used in your model must itself carry an OSI-approved licence that does not limit commercial use, so check the licence of every package and pre-trained model you rely on. Code you share publicly during the competition must go on the competition’s Kaggle forum or notebooks, and is then deemed licensed under an OSI-approved licence that does not limit commercial use. The competition data is licensed CC BY-SA 4.0, yet the Data Security clause says not to share it with anyone who is not participating. Kaggle and the sponsor may use your name and likeness for promotion, and disputes are governed by California law in the courts of Santa Clara County.

Potential winners are notified by email and must reply within one week of the first notification attempt. They then have two weeks from being told which documents are required to return the prize acceptance documents: eligibility certifications, licences and releases, and a US tax form, IRS Form W-9 for US residents or Form W-8BEN for foreign residents, meaning anyone who lives outside the US. Miss either window and the prize goes to the next team. Prizes are paid within about 30 days after Kaggle or the sponsor receives the documents, net of any taxes the sponsor must withhold by law, and winners pay all taxes on their prizes. US residents receive an IRS Form 1099. The rules do not state the payment method or the withholding rate for non-US winners, so check with a tax adviser in your country.

Project ideas that fit the brief

These are our own suggestions, not official tracks. Each one fits the single task in the rules, uses only freely available tools and can run inside an internet-off Kaggle Notebook. Check the licence of any model or package before you build on it.

Tips to give your entry the best chance

• Accept the rules on day one. The entry and team merger cutoff is Wednesday 6 January 2027, 11:59 PM UTC, and joining early costs nothing.
• Get a valid submission.parquet onto the leaderboard in your first session, even with a simple model, so you know the file format, the labels and the internet-off setup all work.
• Build your validation with folds grouped by prompt_id, and weigh that number alongside the public leaderboard rather than trusting the leaderboard alone.
• Before an important submission, time your inference on training passages repeated to about 4,000 rows, so the 9-hour limit never surprises you on the hidden set.
• Add a short check at the end of your notebook that confirms every id appears once, there are no nulls and every label is one of the 18 names, before the file is saved.
• Keep a simple log of every package, dataset and pre-trained model you attach, with its licence. Winners must hand over their code, and an incompatible licence can force rework or disqualification at the winner check.
• Pick your 2 Final Submissions yourself: one that is your best on grouped validation and one that is meaningfully different, rather than leaving the choice to Kaggle.
• Ask important questions on the competition forum, not on Discord, because Kaggle Staff say they do not monitor Discord. Watch the forum for any host answer on generated training data.
• Put both deadlines in your calendar in your own time zone, using the local times in the dates table, so the 11:59 PM UTC cutoff does not catch you on the wrong day.

Please note: we update this page from time to time, but hackathon dates, prizes and rules can change quickly, and there can be a gap between a change and our update. Check the official rules before you enter or submit. This guide helps you plan; it is not legal or tax advice, and ReconScribe is not responsible for decisions made on it.

Name That LLM Kaggle Competition frequently asked questions

What is the Name That LLM Kaggle competition?

It is a Featured Code Competition on Kaggle, sponsored by Google LLC, where you build a model that identifies which of 18 frontier language models wrote each short text passage. You submit a Kaggle Notebook, and Kaggle scores it automatically on about 4,000 hidden passages. Places 1 to 10 on the private leaderboard each win $5,000.

Which countries can enter Name That LLM?

Most of them. The rules open the competition to residents of the United States and worldwide, except residents of Crimea, the so-called Donetsk People’s Republic (DNR), the so-called Luhansk People’s Republic (LNR), Cuba, Iran and North Korea, and anyone subject to US export controls or sanctions. Local laws may still apply, and winners outside the US sign IRS Form W-8BEN as foreign residents.

When is the Name That LLM deadline in my time zone?

Final submissions are due Wednesday 13 January 2027 at 11:59 PM UTC. That is 3:59 pm PST or 6:59 pm EST the same day in the United States, and 12:59 am CET on Thursday 14 January in Central Europe. You must accept the rules and finish any team merger by Wednesday 6 January 2027 at 11:59 PM UTC. The overview countdown shows a close one day later, but plan for 13 January.

How much prize money does Name That LLM pay?

$50,000 in total, paid as ten equal prizes of $5,000 to places 1 to 10 on the private leaderboard. Prizes are paid net of any taxes the sponsor must withhold by law, and winners pay all taxes on their prizes. The competition also awards Kaggle Points and Medals.

Does it cost anything to enter?

No. Entry is free, Kaggle Notebooks are free, and the free GPU quota resets weekly at 30 hours or sometimes more. No paid API or service is required, and no credits are offered for this competition.

Can I upload a CSV of predictions?

No. This is a code competition, so all submissions must come from a Kaggle Notebook and predictions cannot be uploaded directly. Your notebook must write a file named submission.parquet, not a CSV, with the columns id and model.

How is Name That LLM scored?

By balanced accuracy, which is 100% of the score. It averages, over all 18 models, the share of each model’s passages you labelled correctly, so every model counts equally. Final rankings use the private leaderboard, about 60% of the test data, and ties go to the submission entered first.

Can I use pre-trained models or call an AI API?

You can use pre-trained models and external data if they are freely and publicly available, attached to your notebook as Kaggle inputs. Scoring runs have the internet switched off, so your notebook cannot call an online API while it is being scored. Paid tools must meet the rules’ Reasonableness Standard.

Can I generate extra training data with the candidate models?

The rules do not answer this directly. Participants asked on the forum, and as of 17 October 2026 no host reply was visible. The rules only require external data to be publicly available and equally accessible at no cost, or to meet the Reasonableness Standard, so check the competition forum for a host answer before relying on generated data.

How big can a Name That LLM team be?

Up to five people, each with their own Kaggle account. The 5 submissions a day limit applies to the whole team, and the last day to join or merge teams is Wednesday 6 January 2027 at 11:59 PM UTC. Prize money is split evenly unless the team unanimously agrees otherwise and tells Kaggle before payout.

Can students under 18 take part?

Yes, with consent. A pinned Kaggle Staff post says under-18s can take part by submitting two parent or guardian consent forms, one for using Kaggle and one for this competition. Children under 13 are not allowed on Kaggle at all.

What do winners have to hand over?

Winners must deliver training code, inference code, documentation and a description of the computing environment, and license the winning submission under the terms in the rules, which list CC-BY 4.0 and require an OSI-approved licence that does not limit commercial use. The sponsor may also ask for a reproducible write-up with a link to a code repository. Prize acceptance documents, including a W-9 or W-8BEN tax form, are due within two weeks of being told which documents are needed.

More open hackathons

Bolt Builders Week hackathon$17,500Enter by Wed 21 Oct 2026, 06:59 UTCMeta Global AI Developer Hackathon$1,000,000Enter by Mon 26 Oct 2026, 03:59 UTCNebius x NVIDIA Global AI Hackathon$50,000 cash plus 4 NVIDIA Jetson Orin Nano kitsEnter by Fri 30 Oct 2026, 17:00 UTCPayPal AI Hackathon$67,500 cashEnter by Thu 12 Nov 2026, 20:00 UTCMeta VR Start Developer Competition 2026Up to $1,000,000Enter by Wed 18 Nov 2026, 20:00 UTCGemma 4 Developer Agent Competition$100,000Enter by Wed 25 Nov 2026, 23:59 UTCKiroween 2026$100,000 in cashEnter by Fri 4 Dec 2026, 22:00 UTCNFL Big Data Bowl 2027$100,000Enter by Wed 6 Jan 2027, 23:59 UTC

See every open hackathon, with filters and deadlines in your local time

Related reading

• Hackathons 2026: every open hackathon, deadlines in your local time
• Bolt.new free tokens and the Bolt hackathon
• NFL Big Data Bowl 2027: complete guide
• Meta VR Start Developer Competition: complete guide
• Meta AI Developer Hackathon: complete guide
• Free cloud credits for your compute
• AI leaderboard: every major model ranked
• Excel formulas for data work
• Free credits and tools for students

Sources

• Name That LLM competition overview and timeline
• Name That LLM official rules
• Name That LLM Data tab
• Name That LLM leaderboard
• Pinned Kaggle Staff post: how to get started and official Discord
• Kaggle competitions documentation (Code Competition FAQ)
• Kaggle Notebooks documentation
• Kaggle efficient GPU usage documentation