List your company

Spatial decision-making on Maze, MiniBehavior, and FrozenLake Average

79.15Success Rate (%)SpecFlow
36.468447.549258.6369.7108Jun 1, 2026

Evaluation results

MethodLinks
SpecFlowSetting=Flow(image) + AR(Text)2026.0679.15
Gemini-3-FlashSetting=Naive2026.0677.67
DiffThinkerSetting=Flow(image) Only2026.0675.59
GPT-5.1Setting=Naive2026.0674.53
MVoTSetting=Autoregressive2026.0674.1
Qwen3-VL-32BSetting=SFT2026.0663.27
Qwen3-VL-32BSetting=GRPO2026.0659.83
Qwen3-VL-8BSetting=SFT2026.0654.25
Qwen3-VL-8BSetting=GRPO2026.0650.73
Qwen3-VL-32BSetting=Naive2026.0648.02
Qwen3-VL-8BSetting=Naive2026.0638.11