SPIN-Bench - Collaborative: leaderboard
SPIN-Bench cooperative score: the mean of a model's average Hanabi scores in 2-, 3-, 4- and 5-player games, a card game where players see every hand but their own. Points out of 25, higher is better; the project homepage prints the value with a percent sign, but it is on the 0-25 Hanabi scale.
Metric: Mean Hanabi score (0-25). Source: spinbench.github.io. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 | 15.05 | #190 |
| 2 | O4 Mini | 12.4 | #185 |
| 3 | DeepSeek R1 | 11.58 | #262 |
| 4 | O1 Mini | 8.95 | #334 |
| 5 | Claude 3.5 Sonnet | 8.35 | #362 |
| 6 | O3 Mini | 8.3 | #282 |
| 7 | GPT-4 Turbo | 5.45 | #417 |
| 8 | GPT-4o | 5.2 | #357 |
| 9 | Llama 4 Maverick | 4.3 | #486 |
| 10 | Claude 3.5 Haiku | 3 | #571 |
| 11 | Llama 3.3 70B | 1.3 | #578 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spin-bench-collaborative · How It Works · Data refreshed daily, snapshot 2026-11-05.