Atlas

Benchmarks

← All benchmarks

Agents' Last Exam Full-Spectrum Score

Agents · 2026-06-03

Average partial-credit rubric score on the 55-task full-spectrum subset of Agents' Last Exam.

Top models (higher is better)

ModelScore
Claude Opus 5.558.4
GPT-6 Sol53.1
GPT-6 Astra51.8
MiniMax M3.1 Flash50.2
Kimi K349.9
Muse Spark 1.349.2
Claude Opus 547.9
GPT-5.6 Sol46.6
GPT-5.6 Terra46.2
GPT-5.6 Luna43.7
Opus 4.843.1
Claude Fable 542.0
Agnes 2.5 Pro Beta41.5
GPT-5.541.1
GPT-6 Luna40.7
Doubao Seed 2.1 Pro40.1
Opus 4.737.6
GLM-5.237.0
GPT-5.434.3
Composer 2.532.0
Qwen3.7-Max (source unspecified)28.5
Gemini 3.1 Pro Preview26.4
DeepSeek-V4-Pro23.8
Qwen3.6 Plus (2026-04-02)22.9
GLM-5.121.7
MiMo-V2.520.8
Grok 4.318.3
Kimi K2.618.2
MiniMax M2.78.4
Loading Atlas data…