IEEE CIS LLM Domain Q&A Challenge 2026
By IEEE
Check your eligibility
Sign up and we'll tell you whether your startup qualifies for this grant.
Quick Answer
IEEE's online challenge to adapt an open LLM to answer expert questions in two niche fields. US$20,000 in prizes; first place wins US$9,000 (about ₹8.55 lakh).
- Funding amount
- Varies
- Funding type
- Competition
- Provider
- IEEE
- Application deadline
- 30d — 6 Dec 2026
- Location
- International
About this competition
The IEEE CIS Technical Challenge on Domain-Specific Question Answering with LLMs is an online competition organised by the IEEE Computational Intelligence Society Task Force on Large Language Models and Computational Intelligence for General-Purpose AI Systems. Teams adapt a fixed open-weight language model so it can answer expert questions in two specialised areas. Prizes total US$20,000 (about ₹19 lakh), first place wins US$9,000 (about ₹8.55 lakh), and the final submission deadline is 5 December 2026. India is not excluded.
What the program offers
Prizes go to the top five teams: US$9,000 for first (about ₹8.55 lakh), US$5,000 for second, US$3,000 for third, and US$1,500 each for fourth and fifth. The Committee awards prizes based on the final ranking. All teams, whatever their placing, are invited to share a manuscript about their solution and are encouraged to open source their code. This is a contest prize, not funding for a company.
Who can apply
You must be at least 18. The contest is open to residents of the United States and other countries where local law permits participation, and it is void where prohibited. Employees of IEEE's Technical Activities operating unit cannot enter. Individuals or teams of up to four people can take part, each person can be on only one team, and every team member must meet the eligibility rules. Registration needs the organisers' approval.
For an Indian applicant this means you can enter as an individual or a team, provided local law allows it. It suits machine learning engineers, researchers and students with experience in language models, fine-tuning and retrieval. It does not suit someone looking for business funding.
How to apply
Create a Codabench account and request to join the challenge page. Each participant fills in the registration form individually, using the same email as the Codabench account, and the join request is approved after that. Register and form your team of up to four by 15 November 2026. Build your solution with the starter kit on GitHub, which includes a sample of 25 question-answer pairs, starter code and a list of source websites. You may submit once a day. A practical tip: practise on the starter sample first, because your final run must fit strict compute limits.
The task covers two domains: the IFMIF-DONES particle accelerator facility being built in Granada, Spain, and specialised AI research literature. The fixed model is OLMo 3.1 32B Instruct. Your code runs on a single H100 GPU within one hour of total wall-clock time, with no internet access during evaluation, and your files are limited to 1 GB per team. Methods such as parameter-efficient fine-tuning, prompt engineering, retrieval and knowledge graphs are in scope.
How applications are judged and what happens next
Entries are first ranked automatically on the official Codabench leaderboard against a hidden set of expert questions. Shortlisted teams then write a 4-page paper in IEEE format, and a scientific committee and domain experts review the methods and answer quality. That review sets the final ranking, which is the sole basis for the prizes. The dates, all at 11:59 PM UTC, are 15 November 2026 for entry and team mergers, 5 December 2026 for final submissions (5:29 AM IST on 6 December), 10 December 2026 for the first ranking, and 18 December 2026 for the final ranking. Winners are notified by email within 14 days.
About the organisers
The challenge is run by the IEEE CIS GPAIS Task Force, part of the IEEE Computational Intelligence Society, and follows the 2024 FLAME technical challenge. The page shows the Andalusian research institute DaSCI and the IFMIF-DONES Granada project as partner logos.
Things to check before you apply
- Cheating counts. Attempts to manipulate, attack or jailbreak the evaluating model lead to immediate disqualification.
- You must agree to the official rules, which are a separate PDF on the challenge site.
- Runs that exceed the one hour limit are stopped and marked as failed.
- Taxes on any prize are the winner's responsibility, so an Indian winner should check the treatment of foreign prize income with an adviser.
- The source does not say whether any entry fee applies.
Eligibility
See if you're eligible
Full eligibility criteria — entity type, stage, sector, founder profile and registration requirements — plus your fit for this program. Sign up to view.
Organizers
Prizes & Awards
What You Get
IEEE pays US$20,000 (about ₹19 lakh) in prizes to the top five teams: US$9,000 (about ₹8.55 lakh) for first place, US$5,000 for second, US$3,000 for third and US$1,500 each for fourth and fifth. All teams are encouraged to share a write-up and open-source their code after the competition.
- Disbursement
- Lump Sum
- Bootstrap friendly
- Yes
Timelines & Process
Selection Process
Entries are ranked automatically on the Codabench leaderboard. Shortlisted teams then submit a 4-page IEEE-format paper, and a scientific committee and domain experts review the methods and quality of answers. That review sets the final ranking, published on 18 December 2026, which is the sole basis for the prizes. Winners are notified by email within 14 days.
Application Process
- 1
Register
Join on Codabench and fill in the registration form by 15 November 2026.
You - 2
- 3
- 4
- 5
How to apply
Four steps from draft to disbursement — one action each.
- 1
Submit on the official portal
Use the official application link and verify all details before final submission.
- 2
Follow up
Reach out 5–7 days after submitting to confirm receipt and ask about the review timeline.
Frequently Asked Questions
Who can enter the IEEE CIS LLM Domain Q&A Challenge?
People aged 18 or older, alone or in teams of up to four, who are residents of the United States or of other countries where local law permits it. Employees of IEEE's Technical Activities operating unit cannot enter.
Can Indian participants enter the IEEE CIS LLM Domain Q&A Challenge?
Yes. The official rules say the contest is open where local law permits it, and India is not excluded. Registration still needs the organisers' approval.
How much are the IEEE CIS challenge prizes?
Total prizes are US$20,000 (about ₹19 lakh). First place wins US$9,000 (about ₹8.55 lakh), second US$5,000, third US$3,000, and fourth and fifth US$1,500 each.
How are the prizes paid?
The source says taxes on prizes are the winner's responsibility and that winners are notified by email within 14 days. It does not say how or when the money is transferred.
What is the deadline for the IEEE CIS LLM challenge, and what if it has passed?
Register and form teams by 15 November 2026 and submit finally by 5 December 2026, both at 11:59 PM UTC. The source does not say the challenge will repeat, though it follows an earlier 2024 challenge.
Is the IEEE CIS challenge a grant or equity?
It is a prize competition ranked by performance. No equity is taken, and the prizes are not funding for a company.
What do I need to submit for the challenge?
Inference code plus any helper files such as adapters, retrieval indexes or prompt templates, up to 1 GB per team. The final submission also needs a methodology description, and shortlisted teams write a 4-page IEEE-format paper.
Do I need to travel for the IEEE CIS challenge?
No travel is described. The challenge runs online on Codabench.
Is it free to enter the IEEE CIS LLM challenge?
The source does not mention an entry fee. You do need a Codabench account and the organisers' approval.
When are the IEEE CIS challenge results announced?
The first ranking is on 10 December 2026 and the final ranking on 18 December 2026. Winners are notified by email within 14 days.
Sources
Data on this page is compiled from official program and provider references.
