Atlas

Benchmarks

← All benchmarks

Agents' Last Exam Unlicensed Overall Pass Rate

Agents · 2026-06-03

Perfect-score pass rate across the 141 Agents' Last Exam tasks that do not require licensed software, paid APIs, or proprietary data.

Top models (higher is better)

ModelScore
Claude Opus 5.541.1
GPT-6 Astra36.9
Claude Opus 534.8
GPT-6 Sol34.8
Muse Spark 1.334.8
GPT-5.6 Sol33.0
GPT-5.6 Luna32.6
MiniMax M3.1 Flash31.2
GPT-5.6 Terra29.8
Kimi K329.8
Opus 4.829.1
GPT-5.528.7
Claude Fable 527.0
GPT-6 Luna27.0
Agnes 2.5 Pro Beta23.4
Opus 4.722.7
GPT-5.422.1
Composer 2.522.0
GLM-5.222.0
Doubao Seed 2.1 Pro19.9
Gemini 3.1 Pro Preview17.7
DeepSeek-V4-Pro13.4
Qwen3.7-Max (source unspecified)12.8
GLM-5.112.4
Kimi K2.69.9
MiMo-V2.59.2
Qwen3.6 Plus (2026-04-02)9.2
Grok 4.37.8
MiniMax M2.76.4
Loading Atlas data…