Atlas

Benchmarks

← All benchmarks

Agents' Last Exam ALE-CLI Score

Agents · 2026-06-03

Average partial-credit rubric score on the 105-task ALE-CLI Linux-only subset of Agents' Last Exam.

Top models (higher is better)

ModelScore
Claude Opus 5.563.7
GPT-6 Astra61.4
GPT-6 Sol59.5
Muse Spark 1.357.5
Claude Opus 557.3
MiniMax M3.1 Flash53.9
GPT-5.6 Sol53.6
Kimi K352.8
GPT-5.6 Luna52.4
GPT-6 Luna51.9
Agnes 2.5 Pro Beta50.2
GPT-5.6 Terra50.1
GPT-5.549.2
Claude Fable 548.5
Opus 4.745.7
Opus 4.844.6
GLM-5.244.2
Composer 2.542.2
Doubao Seed 2.1 Pro42.2
GPT-5.441.8
Gemini 3.1 Pro Preview37.2
Qwen3.7-Max (source unspecified)34.6
Sonnet 4.632.0
DeepSeek-V4-Pro30.8
GLM-5.130.8
Qwen3.6 Plus (2026-04-02)29.0
MiMo-V2.526.5
Grok 4.325.9
Kimi K2.621.2
MiniMax M2.714.6
Loading Atlas data…