Atlas

Benchmarks

← All benchmarks

Agents' Last Exam ALE-CLI Pass Rate

Agents · 2026-06-03

Perfect-score pass rate on the 105-task ALE-CLI Linux-only subset of Agents' Last Exam.

Top models (higher is better)

ModelScore
Claude Opus 5.534.3
GPT-6 Astra33.3
Muse Spark 1.333.3
GPT-6 Sol32.4
Claude Opus 529.5
GPT-5.6 Luna29.5
Step 5 Preview29.5
GPT-5.6 Sol28.6
GLM-5.328.5
Kimi K327.6
GPT-5.527.1
MiniMax M3.1 Flash26.7
GPT-5.6 Terra26.2
Agnes 2.5 Pro Beta25.7
Opus 4.825.7
Claude Fable 523.8
GLM-5.223.8
Opus 4.722.9
GPT-6 Luna22.9
GPT-5.422.5
Composer 2.519.0
Doubao Seed 2.1 Pro19.0
Gemini 3.1 Pro Preview18.1
DeepSeek-V4-Pro14.1
Sonnet 4.613.3
Qwen3.7-Max (source unspecified)13.3
GLM-5.112.9
Qwen3.6 Plus (2026-04-02)10.5
MiMo-V2.510.0
Grok 4.38.6
Kimi K2.68.1
MiniMax M2.75.7
Loading Atlas data…