Humanity's Last Exam with tools

Tool-augmented Humanity's Last Exam scores reported in DeepSeek-V4 thinking-mode evaluations.

In index
Agentic
Best in agentic

Models36

Top result70.0%Kimi K3.1 Pro

Top-3 spread5.3 pts

Year—

Result and price

22 of 36 results

Results

36 results

#ModelResult

01Kimi K3.1 Pro70.0%

02Claude Opus 5.567.7%

03Claude Opus 564.7%

04Claude Sonnet 5.564.5%

05DeepSeek V4.1 Flash 101763.9%

06LongCat-2.563.1%

07GLM-5.362.5%

08Muse Spark Open61.0%

09DeepSeek V4 Pro 081360.0%

10Hy459.1%

11Claude Haiku 5.557.4%

11Claude Sonnet 557.4%

13GPT-6 Astra57.2%

14Ling 3.0 Flash Sante57.1%

15Qwen3.8 Max56.2%

16Apodex 1.156.1%

16Ornith-1.5-397B56.1%

18Hy4 preview55.4%

19GLM-5.3-Flash55.3%

20Sage-A1.555.1%

21GLM-5.254.7%

22Qwen3.7 Max53.5%

23dots3-note Preview52.6%

24GLM-5.152.3%

25Beam 1.0 501B-A23B51.1%

25Beam 1.0 501B-A23B FP851.1%

25Beam 1.0 501B-A23B NVFP451.1%

28GLM-550.4%

29Agents-A147.6%

30Step 3.7 Flash47.2%

31DeepSeek V4 Flash 073145.1%

32Nemotron 3 Ultra37.4%

33Spark-X2.6-4B36.9%

34Ornith-1.5-35B-A3B33.4%

35Ornith-1.5-9B30.5%

36LFM3-1.7B13.9%