Atlas

Benchmarks

← All benchmarks

Agents' Last Exam Overall Score

Agents · 2026-06-03

Average partial-credit rubric score across the full 152-task Agents' Last Exam set, reported separately from the stricter perfect-score pass rate.

Top models (higher is better)

ModelScore
Claude Opus 5.563.2
GPT-6 Astra59.3
GPT-6 Sol58.6
Claude Opus 555.9
Muse Spark 1.355.8
MiniMax M3.1 Flash55.0
GPT-5.6 Sol53.6
Kimi K351.6
GPT-6 Luna50.9
GPT-5.6 Terra50.7
GPT-5.6 Luna50.4
Claude Fable 548.7
GPT-5.547.9
Opus 4.845.2
Agnes 2.5 Pro Beta42.7
Opus 4.742.5
Doubao Seed 2.1 Pro41.9
GLM-5.241.1
Composer 2.539.5
GPT-5.437.3
Gemini 3.1 Pro Preview32.7
Qwen3.7-Max (source unspecified)31.6
GLM-5.128.1
DeepSeek-V4-Pro27.6
Qwen3.6 Plus (2026-04-02)24.6
MiMo-V2.523.6
Kimi K2.621.7
Grok 4.321.3
MiniMax M2.714.2
Loading Atlas data…