Open-ended Question Answering on PhysReason v2 (test)
51.1Subpart-AND (v2)GPT-4o
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-4o2026.05 | 51.1 | |
| Claude Sonnet 4.5max_tokens=163842026.05 | 49.1 | |
| Physics-R1 (binary, 3-seed mean ±σ)max_tokens=16384, reward_type=binary, aggregation=3-seed mean2026.05 | 39.6 | |
| Gemini 2.5 Pro2026.05 | 38.8 | |
| Physics-R1 (binary, seed 42)max_tokens=16384, reward_type=binary, seed=422026.05 | 32.2 | |
| Qwen3-VL-32B-Thinking2026.05 | 25.1 | |
| Qwen3-VL-8B-Thinking (base)2026.05 | 23.9 | |
| Physics-R1 (dense)max_tokens=16384, reward_type=dense2026.05 | 23.3 | |
| InternVL3-8B2026.05 | 13.3 |