#1 on WebBench

AgentSmith achieves 82.4% success rate on the Halluminate WebBench — the industry standard for evaluating browser agents across 5,000+ real-world tasks.

February 2026 · View benchmark on GitHub

AgentSmith
82.4%
rtrvr.ai
81.4%
Operator + Human
76.5%
Anthropic CUA
66.0%
Skyvern 2.0
64.4%
OpenAI Operator
59.8%
Browser Use Cloud
43.9%
Convergence AI
39.9%
AgentSmith
89.1%
rtrvr.ai
88.2%
Anthropic CUA
80.6%
Operator + Human
79.0%
Skyvern 2.0
74.2%
OpenAI Operator
75.0%
Browser Use Cloud
63.2%
Convergence AI
51.8%
AgentSmith
68.7%
Operator + Human
70.7%
rtrvr.ai
65.6%
Skyvern 2.0
46.6%
Anthropic CUA
39.4%
OpenAI Operator
32.3%
Convergence AI
13.1%
Browser Use Cloud
11.4%

Methodology

Results from the Halluminate WebBench — 5,294 tasks across 350+ live websites in five categories (READ, CREATE, UPDATE, DELETE, FILE_MANIPULATION). Each task is evaluated by an LLM judge. Competitor scores sourced from their published results. AgentSmith tested under the same conditions with no task-specific tuning.

Install AgentSmith and see the results firsthand. 200 actions/day, free forever.

Install for Chrome – Free