Skip to content

Benchmark

WebArena

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
9
Scores recorded
12
Sources
2
IQ weight
1.20

Score distribution

45.4%
Mean
58.9%
Median
27.3%
Std dev
7.1%
10th percentile
71.2%
90th percentile
1.4%
Min
1.4%72.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.20 for chat models, measured over 9 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 4.5pass@4 · Computer Use API + browser tools · single policy · general prompts · official grader· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.72.4%
2Claude Opus 4.5pass@3 · Computer Use API + browser tools · single policy · general prompts · official grader· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.71.2%
3Claude Opus 4.5pass@2 · Computer Use API + browser tools · single policy · general prompts · official grader· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.69.5%
4Claude Sonnet 4.6single policy · screenshot + DOM navigation · general prompts · Average@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.65.6%
5Claude Opus 4.5pass@1 · Computer Use API + browser tools · single policy · general prompts · official grader · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.65.3%
6Kimi K2.5· Steel.dev Agent Leaderboards data58.9%
7Claude Sonnet 4.5pass@1 · Computer Use API + browser tools · single policy · general prompts · official grader · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.58.5%
8Claude Haiku 4.5pass@1 · Computer Use API + browser tools · single policy · general prompts · official grader · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53.1%
9GPT 4o· Steel.dev Agent Leaderboards data13.1%
10GPT 35 Turbo 16k· Steel.dev Agent Leaderboards data8.9%
11Gemini Pro· Steel.dev Agent Leaderboards data7.1%
12mixtral-8x7b· Steel.dev Agent Leaderboards data1.4%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Steel.dev Agent Leaderboards, MIT (licence)