List your company

Multiple-Choice Question Answering on PhyX 1k (test)

79.7MCQ Exact AccuracyClaude Sonnet 4.5
45.48454.36763.2572.133May 13, 2026

Evaluation results

MethodLinks
Claude Sonnet 4.5max_tokens=163842026.0579.7
Physics-R1 (dense)max_tokens=16384, reward_type=dense2026.0578.3
Physics-R1 (binary, seed 42)max_tokens=16384, reward_type=binary, seed=422026.0578
Physics-R1 (binary, 3-seed mean ±σ)max_tokens=16384, reward_type=binary, aggregation=3-seed mean2026.0577.8
Gemini 2.5 Promeasured_by=authors2026.0575.1
Qwen3-VL-32B-Thinking2026.0573.8
Qwen3-VL-8B-Thinking (base)2026.0573.7
GPT-4osource=Shen et al. [2025]2026.0570.4
InternVL3-8B2026.0546.8