Home / Competitions
Competitions
Auto-graded, head-to-head AI benchmarks with real cash prizes. Live leaderboards. Open to anyone with a working submission — no application, no CV, no institutional affiliation needed.
Live
Active competitions
Submissions are scored against a private held-out eval. Leaderboards update every 15 minutes. Final ranks lock at 23:59 UTC on the deadline. Prize money is wired within 14 days of close.
Archive
Past competitions
Every competition is archived with the winner's submission code, the eval-set spec, the held-out test set (released 30 days after close), and the full final leaderboard.
| Year | Competition | Prize | Winner | Runners-up |
|---|---|---|---|---|
| 2026 | Cohort-5 RAG Re-rank | $10,000 | minou (FR) | henrik.l, sofia_k |
| 2025 | Long-Context KV Pareto | $18,000 | tomek_w (PL) | kenji, priya.a |
| 2025 | Agent Schema Fuzz | $9,000 | kenji (JP) | dineth.k, danielb |
| 2025 | Indic-Eval Harness Sprint | $6,500 | vihaan (IN) | noor.r, aditi.r |
| 2025 | Refusal Direction Reproduction | $4,000 | sofia_k (GR) | anya.p, kari.n |
| 2024 | Cantonese Tool-Use Bench | $8,500 | wai.lin (HK) | loke.w, hwong |
How they work
Four steps from idea to prize
Pick a competition
Each has a one-page spec, a held-out eval, and a starter notebook.
Submit code
You submit code, not predictions. The grader runs it in a sandboxed container.
Iterate publicly
Leaderboard is public. Submit as often as you like; it updates every 15 minutes.
Win, get paid
Top three get cash. Everyone who clears the baseline gets contributor points.