Long-horizon task solving on PINCHBENCH 123 tasks expanded
90.3Best Success RateClaude-Haiku-4.5
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Claude-Haiku-4.5Source type=Closed-source models2026.06 | 90.3 | 58.9 | |
| Claude-Opus-4.6Source type=Closed-source models2026.06 | 88.9 | 67.5 | |
| Qwen3.6-PlusSource type=Open-source / open-weight models2026.06 | 82.8 | 62.6 | |
| Gemini-3.1-Pro-PreviewSource type=Closed-source models2026.06 | 81.7 | 80.2 | |
| OpenClaw-Skill 9BSource type=Open-source / open-weight models2026.06 | 68.2 | 53.6 | |
| OpenClaw-Skill 4BSource type=Open-source / open-weight models2026.06 | 61.4 | 47.6 | |
| Qwen3.5-9BSource type=Open-source / open-weight models2026.06 | 61.1 | 47.1 | |
| Qwen3.5-4BSource type=Open-source / open-weight models2026.06 | 60.9 | 45.9 | |
| Nemotron-3-Super-120B-A12BSource type=Open-source / open-weight models2026.06 | 50.6 | 38.6 | |
| GPT-OSS-120BSource type=Open-source / open-weight models2026.06 | 47.4 | 44.5 | |
| GPT-OSS-20BSource type=Open-source / open-weight models2026.06 | 41.8 | 37.5 | |
| OpenClaw-Skill-Qwen3 8BSource type=Open-source / open-weight models2026.06 | 33 | 22.5 | |
| OpenClaw-Skill-Qwen3-4BSource type=Open-source / open-weight models2026.06 | 31.1 | 20.8 | |
| Qwen3-8BSource type=Open-source / open-weight models2026.06 | 27.9 | 18.3 | |
| Qwen3-4BSource type=Open-source / open-weight models2026.06 | 22.4 | 13.6 |