Atlas

Benchmarks

← All benchmarks

Agents' Last Exam Unlicensed Near-Term Score

Agents · 2026-06-03

Average partial-credit rubric score on the 67-task unlicensed near-term subset of Agents' Last Exam.

Top models (higher is better)

ModelScore
Claude Opus 5.583.9
GPT-6 Astra82.3
GPT-6 Sol80.9
Claude Opus 580.1
GPT-5.6 Sol78.8
Muse Spark 1.377.9
MiniMax M3.1 Flash77.2
GPT-5.6 Luna75.6
GPT-5.6 Terra73.9
GPT-6 Luna73.4
GPT-5.572.4
Kimi K371.6
Claude Fable 571.1
Opus 4.864.0
Opus 4.762.7
Composer 2.562.3
GLM-5.260.3
Agnes 2.5 Pro Beta60.2
Doubao Seed 2.1 Pro60.1
GPT-5.457.8
Gemini 3.1 Pro Preview54.9
Qwen3.7-Max (source unspecified)46.9
GLM-5.145.6
DeepSeek-V4-Pro43.8
Qwen3.6 Plus (2026-04-02)36.4
Kimi K2.635.1
MiMo-V2.535.1
Grok 4.331.8
MiniMax M2.724.5
Loading Atlas data…