CL-Bench Codebase Adaptation Gain
Agents · 2026-05-04
Mean cumulative gain over the stateless baseline across 19 sequential instances of the CL-Bench codebase adaptation task.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.4 | 3.0 |
| Sonnet 4.6 | 2.7 |
| Opus 4.7 | 1.5 |
| Gemini 3 Flash Preview | -0.3 |
| Gemini 3.1 Pro Preview | -1.3 |