List your company

Windows UI Navigation on WindowsAgentArena (WAA)

74.5Success RateHuman Performance
3.67622.06340.4558.837Jan 28, 2026Feb 17, 2026Mar 9, 2026Mar 29, 2026Apr 18, 2026May 8, 2026May 28, 2026

Evaluation results

MethodLinks
Human Performance2026.0174.5—
CUA-Skill AgentBackbone=GPT-5, Decoding Strategy=Best of 32026.0157.530
AgentS3Backbone=GPT-5, Decoding Strategy=Best-of-32026.0156.6100
CUA-Skill AgentBackbone=GPT-52026.0150.330
AgentS3Backbone=GPT-52026.014950
RoTS-32BMax Steps=502026.0539.1—
Operator2026.0137.450
RoTS-32BMax Steps=152026.0535.9—
Jedi-7B w/ GPT4-oMax Steps=502026.0532.9—
UFO-2Backbone=GPT-o12026.0130.530
Jedi-7B w/ GPT4-oMax Steps=152026.0530.2—
Agent-S2Backbone=Claude-3.5-Sonnet2026.0129.815
RoTS-7BMax Steps=502026.0528.2—
RoTS-7BMax Steps=152026.0524.9—
ScaleCUA-32BMax Steps=502026.0524.2—
ScaleCUA-32BMax Steps=152026.0521.4—
Ultra-CUA-7BBackbone=QWen2.5VL-FT2026.0121.115
ScaleCUA-7BMax Steps=502026.0520.7—
STEVEBackbone=GPT4o2026.0120.140
NAVIBackbone=GPT4o2026.0119.515
Agent-SBackbone=GPT-4o2026.0118.215
UITARS-1.5-7BBackbone=Qwen2.5VL-FT2026.0118.115
ScaleCUA-7BMax Steps=152026.0518—
UI-TARS-72B-DPOMax Steps=502026.0517.9—
UI-TARS-1.5-7BMax Steps=502026.0515.9—
UI-TARS1Backbone=Qwen2VL-FT2026.0115.750
OpenCUA-7BMax Steps=152026.0513.5—
Qwen2.5-VL-72BMax Steps=152026.0511.8—
UI-TARS-1.5-7BMax Steps=152026.0511.1—
UI-TARS-72B-DPOMax Steps=152026.0511.1—
Qwen2.5-VL-72BMax Steps=502026.059.7—
Claude 3.7 SonnetMax Steps=152026.057.1—
Claude 3.7 SonnetMax Steps=502026.056.4—