Software Engineering on SWE-Bench Verified (SWE-Agent Metrics)
78.2SWE-Agent ScoreClaude-Opus-4.5
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Claude-Opus-4.5Size=?, Maximum agent turns=3002026.02 | 78.2 | 77.8 | 79 | |
| Claude-Sonnet-4.5Size=?, Maximum agent turns=3002026.02 | 76 | 68.4 | 74.6 | |
| MiniMax-M2.1Size=230A10, Maximum agent turns=3002026.02 | 74.8 | 70.4 | 71 | |
| GLM-4.7Size=358A32, Maximum agent turns=3002026.02 | 74.2 | 70.4 | 70.6 | |
| Kimi-K2.5Size=1000A32, Maximum agent turns=3002026.02 | 73.2 | 70.8 | — | |
| Qwen3-Coder-NextSize=80A3, Maximum agent turns=3002026.02 | 70.6 | 71.1 | 71.3 | |
| DeepSeek-V3.2Size=671A37, Maximum agent turns=3002026.02 | 70.2 | 67.2 | 72.6 |