Humanity's Last Exam (Self-Reported, With Tools): leaderboard

Metric: Accuracy (%). Source: huggingface.co. 26 models tracked.

Top models

#ModelScore
1GLM-5.362.5
2Hy4 preview55.4
3GLM-5.3 Flash55.3
4GLM-5.254.7
5Kimi K2.654
6GLM-5.152.3
7GLM-550.4
8Kimi K2.550.2
9Qwen 3.5 27B48.5
10Qwen 3.5 397B A17B48.3
11Step 3.7 Flash48.1
12MiMo-V2.5-Pro48
13Qwen 3.5 122B A10B47.5
14Kimi K2 (Thinking)44.9
15GLM-4.742.8

Interactive version: theaggregate.ai/benchmark?slug=humanity-s-last-exam-self-reported-with-tools · How It Works · Data refreshed daily, snapshot 2026-09-05.