CL-Bench Exploitable Poker Gain
Agents · 2026-05-04
Mean cumulative gain over the stateless baseline across 120 sequential instances of the CL-Bench exploitable poker task.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 4.6 | 58.5 |
| Gemini 3.1 Pro Preview | 32.9 |
| GPT-5.4 | 20.5 |
| Opus 4.7 | -41.0 |
| Gemini 3 Flash Preview | -102 |