List your company

Cross-task physical reasoning on PHYRE Set 1 (test)

52Pass@1Human
-1.87212.11426.140.086Jul 7, 2026

Evaluation results

MethodLinks
HumanCategory=Oracle2026.075275.880
+EG+VAORABase Model=Qwen3VL-8B-SFT, Protocol=Expert-Guided + VAORA2026.0738.252.459.4
DQN-expertCategory=Oracle2026.0732.643.447.6
Gemini-3.1-ProCategory=Closed-Source API2026.0727.847.255.2
+EG (Expert-Guided)Base Model=Qwen3VL-8B-SFT, Protocol=Expert-Guided2026.072743.651.2
Qwen3VL-8B (SFT)Category=Open-Sourced Baselines, Protocol=SFT2026.0717.63847
Gemini-3.1-FlashCategory=Closed-Source API2026.07173441.6
GPT-5.4Category=Closed-Source API2026.076.413.218.6
InternVL-3.5-8BCategory=Open-Sourced Baselines2026.074.49.413.4
Claude-Sonnet-4.6Category=Closed-Source API2026.074.210.414.2
Qwen3VL-8B (SFT+GRPO)Category=Open-Sourced Baselines, Protocol=SFT+GRPO2026.0722.63
Qwen3VL-8B (GRPO)Category=Open-Sourced Baselines, Protocol=GRPO2026.070.646.2
Qwen3VL-8BCategory=Open-Sourced Baselines2026.070.20.820