Early-stopping budget estimation on Search-R1, Sokoban, SWE-bench, and Warehouse Aggregate (test)
2.2False-Abort RateClaude Opus 4.7
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Claude Opus 4.7Model=Claude Opus 4.72026.05 | 2.2 | 28.2 | 50 | 62 | |
| Gemini 3.1 ProModel=Gemini 3.1 Pro2026.05 | 2.8 | 55.7 | 63 | 123 | |
| Claude Sonnet 4.6Model=Claude Sonnet 4.62026.05 | 3.3 | 49.6 | 76 | 101 | |
| Qwen3 235BModel=Qwen3 235B2026.05 | 4.9 | 38.8 | 190 | 140 | |
| GPT-5.2 InstantModel=GPT-5.2 Instant2026.05 | 6.6 | 64.1 | 183 | 124 |