CL-Bench Sales Prediction Reward
Agents · 2026-05-04
Mean cumulative reward across 12 sequential instances of the CL-Bench sales prediction task.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 4.6 | 10.1 |
| GPT-5.4 | 9.2 |
| Opus 4.7 | 9.0 |
| Gemini 3 Flash Preview | 8.5 |
| Gemini 3.1 Pro Preview | 8.1 |