Databricks is hosting a live competition at the Data + AI Summit 2026, where student teams will build AI agents for enterprise reasoning tasks. Full details on the competition and prizes are included below.
Overview
The Grounded Reasoning Cup is a live competition bringing together leading AI labs and top academic teams to tackle real-world enterprise challenges using AI agents.
The competition is built around OfficeQA, a benchmark developed by Databricks that evaluates how well AI systems can retrieve and reason over large collections of real-world unstructured documents in enterprise settings. Teams will have 2 months prior to the competition to build out their solutions, and will apply their agents to a new, unseen test set in the competition—testing how well solutions generalize.
This is a unique opportunity to:
- Tackle high-impact, real-world enterprise problems with AI
- Showcase your work in front of top researchers, engineers, and industry leaders
- Compete for recognition—and $100k in model credit prizes—at one of the largest data & AI conferences
Application
The application window is now closed. Thank you to all teams who applied!
Competition Format & Rules
Development Period
The main development period is April 15 – June 15. During this time, you will develop and refine your approach with the existing OfficeQA benchmark serving as a development set.
Industry Lab Sponsors
Each team will be paired with a single frontier model provider as their sponsor. We are excited to be working with OpenAI, Anthropic, and Google as sponsors for the Grounded Reasoning Cup.
Each sponsor will provide weekly mentorship and general support to help their paired team develop state-of-the-art agents built on the sponsor’s models, and optionally their tooling (e.g., agent SDKs), to be used for the competition.
Live Competition
At the live event, teams will be evaluated on a previously unseen benchmark built from data similar to the U.S. Treasury Bulletins (the original corpus used for OfficeQA).
| Detail | Information |
|---|---|
| Date | June 17, 2026 |
| Time | 1:00 – 4:00 PM |
| Location | Moscone Center West, San Francisco |
Competition Flow
The competition consists of 6 rounds over approximately two hours.
- Rounds: In each round, 10 new questions from the benchmark are released simultaneously to all teams. Teams have 15 minutes to submit answers to all 10 questions.
- Systems: You may use AI agents, automated pipelines, or human-in-the-loop review.
- Re-Submissions: Each team has 3 re-submit opportunities throughout the competition to resubmit an answer in the current round.
Scoring
- Correct Answer: Each correct answer earns 1 point.
- Speed Bonus: The first team to correctly answer a given question receives a +0.25 pt speed bonus.
- Final Ranking: Determined by total points accumulated.
Prizes
The top three academic teams will receive awards on stage at DAIS, as well as $100k worth of Databricks credits:
Credits can be used to serve models from OpenAI, Anthropic, Google, and more.
Timeline Summary
| Phase | Dates |
|---|---|
| Team applications (closed) | Mar 23 – Apr 5 |
| Application review & selection | Apr 5 – Apr 13 |
| Development period | Apr 15 – Jun 15 |
| Competition | Jun 17 (1–4 PM) |