CUA-bench — Hidden Reflex
Agents · 2026-09-18
Ordered-milestone progress on the held-out reflex game within three hours.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 22.0 |
| Claude Opus 5 | 20.0 |
| Gemini 3.8 Flash | 20.0 |
| GPT-6 Astra | 20.0 |
| Claude Fable 5.1 | 12.0 |