Atlas

Benchmarks

← All benchmarks

Agents' Last Exam Unlicensed Full-Spectrum Score

Agents · 2026-06-03

Average partial-credit rubric score on the 50-task unlicensed full-spectrum subset of Agents' Last Exam.

Top models (higher is better)

ModelScore
Claude Opus 5.558.5
GPT-6 Sol54.6
GPT-6 Astra51.6
Muse Spark 1.350.3
MiniMax M3.1 Flash49.5
Claude Opus 548.8
Kimi K348.1
GPT-5.6 Sol46.3
GPT-5.6 Terra45.1
GPT-5.6 Luna44.3
Agnes 2.5 Pro Beta43.7
Opus 4.841.9
GPT-6 Luna40.6
Claude Fable 540.4
GPT-5.539.4
Opus 4.739.3
Doubao Seed 2.1 Pro39.0
GLM-5.236.8
GPT-5.434.9
Composer 2.531.9
Qwen3.7-Max (source unspecified)30.4
Gemini 3.1 Pro Preview25.7
DeepSeek-V4-Pro24.9
Qwen3.6 Plus (2026-04-02)24.8
GLM-5.123.0
MiMo-V2.521.7
Grok 4.319.9
Kimi K2.618.8
MiniMax M2.78.2
Loading Atlas data…