Atlas

Benchmarks

← All benchmarks

Agents' Last Exam Full-Spectrum Pass Rate

Agents · 2026-06-03

Perfect-score pass rate on the 55-task full-spectrum subset, which provides broad coverage across more than 50 professional subdomains.

Top models (higher is better)

ModelScore
Claude Opus 5.532.7
GPT-6 Astra30.9
GPT-5.6 Sol30.0
Claude Opus 529.1
GPT-5.6 Luna29.1
GPT-6 Sol29.1
Kimi K329.1
Muse Spark 1.329.1
GPT-5.6 Terra27.3
Claude Fable 525.5
MiniMax M3.1 Flash25.5
Doubao Seed 2.1 Pro24.1
Agnes 2.5 Pro Beta23.6
Opus 4.823.6
GPT-5.523.6
GPT-6 Luna23.6
Opus 4.720.0
GPT-5.419.4
Composer 2.518.2
GLM-5.218.2
Gemini 3.1 Pro Preview12.7
DeepSeek-V4-Pro10.9
Qwen3.7-Max (source unspecified)10.9
GLM-5.19.1
MiMo-V2.59.1
Qwen3.6 Plus (2026-04-02)8.2
Grok 4.37.3
Kimi K2.66.4
MiniMax M2.73.6
Loading Atlas data…