Atlas

Benchmarks

← All benchmarks

OSWorld (self-reported, A11y tree)

Agents · 2024-04-11

Self-reported OSWorld success rate with accessibility-tree observation.

Top models (higher is better)

ModelScore
Claude 3.5 Sonnet (June 2024)22.5
GPT-4o16.6
GPT-412.2
GPT-4 Turbo10.8
Gemini 1.5 Pro10.3
Qwen Max (source-unspecified mainline)6.9
GPT-4o Mini6.2
Qwen VL Max 08094.3
Mixtral 8x7B3.0
LLaVA-OneVision 72B2.7
Gemini 1.0 Pro2.4
Llama 3 70B1.6
Loading Atlas data…