TeamGym - EvidenceBench - Single-Agent All-PIU: leaderboard
TeamGym measures how agents combine information to solve tasks. Results distinguish individual agents, cooperating agents and how much information each receives. Episode success (%).
Metric: Episode success (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 93.3 | #32 |
| 2 | Qwen 3 235B A22B | 62.2 | #317 |
| 3 | Claude Sonnet 4.5 | 57.8 | #146 |
| 4 | Claude Haiku 4.5 | 53.3 | #283 |
| 5 | Gemma 3 27B (IT) | 44.4 | #521 |
| 6 | Qwen 3 32B | 37.8 | #443 |
| 7 | Gemma 3 12B (IT) | 33.3 | #662 |
| 8 | Gemma 3 4B (IT) | 13.3 | #1010 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=teamgym-evidencebench-single-agent-all-piu · How It Works · Data refreshed daily, snapshot 2026-10-29.