Open-ended Question Answering on PHYSOLYM-A v1 (held-out)
33.4Problem-level ScoreClaude Sonnet 4.5
Evaluation results
| Method | Links | |
|---|---|---|
| Claude Sonnet 4.5max_tokens=163842026.05 | 33.4 | |
| Physics-R1 (binary, 3-seed mean ±σ)max_tokens=16384, reward_type=binary, aggregation=3-seed mean2026.05 | 26.3 | |
| Physics-R1 (binary, seed 42)max_tokens=16384, reward_type=binary, seed=422026.05 | 25.6 | |
| GPT-4o2026.05 | 19.5 | |
| Physics-R1 (dense)max_tokens=16384, reward_type=dense2026.05 | 19.2 | |
| Qwen3-VL-32B-Thinking2026.05 | 13.2 | |
| Gemini 2.5 Pro2026.05 | 12.2 | |
| Qwen3-VL-8B-Thinking (base)2026.05 | 8 | |
| InternVL3-8B2026.05 | 4 |