List your company
ResearchFormerly WizWand

Visual Agentic Reasoning on Sokoban

85Success RateGLANCE-Full
10.1229.564968.44Mar 19, 2026Mar 26, 2026Apr 3, 2026Apr 11, 2026Apr 19, 2026Apr 27, 2026May 5, 2026
Updated 6mo ago

Evaluation results

MethodLinks
GLANCE-FullBackbone=Qwen2.5-VL-3B, Reward Strategy=Dense Extrinsic, Reasoning Strategy=World Model Reasoning for Visual States2026.0585—
VAGEN-FullBackbone=Qwen2.5-VL-3B, Reward Strategy=Dense Extrinsic, Reasoning Strategy=World Model Reasoning for Visual States2026.0579—
GLANCE-BaseBackbone=Qwen2.5-VL-3B, Reward Strategy=Sparse Extrinsic, Reasoning Strategy=World Model Reasoning Strategy2026.0574—
RewardFlowType=RL Training, Backbone=Qwen2.5-(VL)-7B-Instruct2026.0362.43
VAGEN-BaseBackbone=Qwen2.5-VL-3B, Reward Strategy=Sparse Extrinsic, Reasoning Strategy=World Model Reasoning Strategy2026.0561—
Gemini 2.5 Pro2026.0558—
GLANCE w/ Turn-PPOBackbone=Qwen2.5-VL-3B, Reward Strategy=Turn-level PPO, Reasoning Strategy=World Model Reasoning Strategy2026.0552—
RewardFlowType=RL Training, Backbone=Qwen2.5-(VL)-3B-Instruct2026.0349.22.2
o4-mini2026.0544—
GPT-4o2026.0543—
Turn-PPO w/ MaskBackbone=Qwen2.5-VL-3B, Reward Strategy=Turn-level PPO, Reasoning Strategy=World Model Reasoning Strategy2026.0538—
GiGPOType=RL Training, Backbone=Qwen2.5-(VL)-7B-Instruct2026.0334.41.4
Claude 4.5 Sonnet2026.0531—
GRPOType=RL Training, Backbone=Qwen2.5-(VL)-3B-Instruct2026.0326.61.3
Claude 3.7 Sonnet2026.0525—
GRPOType=RL Training, Backbone=Qwen2.5-(VL)-7B-Instruct2026.0323.41
RLOOType=RL Training, Backbone=Qwen2.5-(VL)-3B-Instruct2026.0322.71
GiGPOType=RL Training, Backbone=Qwen2.5-(VL)-3B-Instruct2026.0321.91.2
RLOOType=RL Training, Backbone=Qwen2.5-(VL)-7B-Instruct2026.0321.91
Qwen2.5-VL-72BBackbone=Qwen2.5-VL-72B2026.0520—
GRPO w/ MaskBackbone=Qwen2.5-VL-3B, Reward Strategy=RL Baseline, Reasoning Strategy=World Model Reasoning Strategy2026.0520—
BaseType=Prompting, Backbone=Qwen2.5-(VL)-7B-Instruct2026.0318.80.9
Vanilla-PPOBackbone=Qwen2.5-VL-3B, Reward Strategy=RL Baseline, Reasoning Strategy=World Model Reasoning Strategy2026.0518—
VLM-R1-3BBackbone=VLM-R1-3B2026.0516—
BaseType=Prompting, Backbone=Qwen2.5-(VL)-3B-Instruct2026.0314.10.5
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.0514—
Qwen2.5-VL-3BBackbone=Qwen2.5-VL-3B2026.0513—