List your company

Visual Web Navigation on WebGym OOD 2026 (test)

52.4Peak Success Rate (Easy)AsyncWebRL (full)
31.70437.07742.4547.823Jun 4, 2026

Evaluation results

MethodLinks
AsyncWebRL (full)Model=Qwen3-VL-8B-Instruct2026.0652.434.37.145.4
AsyncWebRL (full)Model=Qwen3-VL-8B-Thinking2026.0651.835.111.344.4
WebGym (sync REINFORCE)Model=Qwen3-VL-8B-Instruct2026.0650.924.14.842.9
AsyncWebRL-RAFT++Model=Qwen3-VL-8B-Thinking2026.0647.3305.240.5
AsyncWebRL-RAFT++Model=Qwen3-VL-8B-Instruct2026.0646.627.85.539.3
Base (no RL)Model=Qwen3-VL-8B-Thinking2026.0637.424.31.232
Base (no RL)Model=Qwen3-VL-8B-Instruct2026.0632.511.2026.2