Atlas

Benchmarks

← All benchmarks

Agents' Last Exam Unlicensed Last-Exam Score

Agents · 2026-06-03

Average partial-credit rubric score on the 30-task unlicensed Last-Exam subset of Agents' Last Exam.

Top models (higher is better)

ModelScore
Claude Opus 5.536.4
GPT-6 Astra32.0
GPT-6 Sol31.3
Claude Opus 530.4
GPT-6 Luna29.7
Muse Spark 1.326.9
Kimi K323.1
GPT-5.6 Sol20.7
MiniMax M3.1 Flash20.7
Claude Fable 516.4
Opus 4.816.1
Agnes 2.5 Pro Beta15.9
GPT-5.6 Luna15.1
GPT-5.515.0
GPT-5.6 Terra15.0
GLM-5.213.0
Opus 4.712.8
Composer 2.511.1
GPT-5.410.3
Doubao Seed 2.1 Pro9.1
Qwen3.7-Max (source unspecified)8.1
GLM-5.16.2
Qwen3.6 Plus (2026-04-02)4.6
MiMo-V2.54.3
Gemini 3.1 Pro Preview3.9
DeepSeek-V4-Pro3.2
Grok 4.33.0
Kimi K2.62.1
MiniMax M2.71.6
Loading Atlas data…