Atlas

Benchmarks

← All benchmarks

Agents' Last Exam - release-reported pass rate, subset unspecified

Agents

Release-reported perfect-task pass rate with no established Linux105 versus full152 task cohort. Diagnostic until the subset is identified; distinct from Sakana's AtCoder ALE-Bench.

Top models (higher is better)

ModelScore
DeepSeek V4.1 Flash 101735.4
DeepSeek V4.1 Flash31.8
MiMo-V2.6-Pro31.6
MiMo-V2.6-Flash27.6
Gemini 3.7 Flash26.3
DeepSeek V4 Pro 081325.7
Qwen3.8-Flash-Next24.3
Loading Atlas data…