Frontier Log

Benchmark · Active

Humanity's Last Exam

Created by Center for AI Safety and Scale AI, 23 Jan 2025 · Higher is better · Website

0204060Oct 2025Jan 2026Apr 2026Jul 2026Oct 202651.7%Qwen4-Max36.5%Qwen3.8-Omni-Flash43.9%Qwen3.8-Max-090234.7%Qwen3.7-Plus21.4%Qwen3.6-35B-A3B19.8%DeepSeek-V3.2-Exp13.6%Qwen3-VL-235B-A22B-Thinking0204060Jan 2026Jul 202651.7%Qwen4-Max43.9%Qwen3.8-Max-090234.7%Qwen3.7-Plus21.4%Qwen3.6-35B-A3B19.8%DeepSeek-V3.2-Exp

Settings and run setups

Task setToolsProviders
Full set (2,500 questions)No toolsOpenAIAnthropicGoogleMetaMoonshot AI
Full set (2,500 questions)With toolsOpenAIAnthropicGoogleMetaMoonshot AIZ.aiMeituanTencent
Text-only subsetNo toolsOpenAIMetaDeepSeekAlibabaMoonshot AIMiniMaxZ.aiNVIDIATencentThinking Machines
Text-only subsetWith toolsDeepSeekMoonshot AIZ.aiNVIDIATencent
Not statedNo toolsOpenAIGoogleDeepSeekAlibabaMiniMaxZ.aiNVIDIA
Not statedWith toolsOpenAIDeepSeekAlibabaZ.aiNVIDIAThinking Machines
Full set (2,500 questions)Not statedNVIDIA
Text-only subsetNot statedThinking Machines
Not statedNot statedDeepSeekAlibaba

Download image

Results

ModelScoreSource
Qwen4-Max16 Nov 202651.7%Qwen4-Max: Your New Teammate Has ArrivedSource image
Qwen3.8-Max-09022 Sep 202643.9%Qwen4-Max: Your New Teammate Has Arrivedreported laterSource image
Qwen3.8-Omni-Flash18 Sep 202636.5%Qwen3.8-Omni-Flash: Omni Senses. Agentic Delivery.Checked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.GPT-4o judgeOur capture
Qwen3.8-Flash-Next26 Aug 202635.9%Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-EfficiencyCorroborated The same number is printed in 1 other lab document: 1GPT-4o judgeOur capture
Qwen3.7-Plus1 Jun 202634.7%Qwen3.7-Plus: Multimodal Agent IntelligenceChecked A person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Corroborated The same number is printed in 1 other lab document: 1xhigh reasoning effortOur capture
Qwen3.6-35B-A3B15 Apr 202621.4%Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to AllChecked A person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Corroborated The same number is printed in 1 other lab document: 1Our capture
DeepSeek-V3.2-Exp29 Sep 202519.8%Introducing DeepSeek-V3.2-ExpCorroborated The same number is printed in 1 other lab document: 1Source image
Qwen3-VL-235B-A22B-Thinking23 Sep 202513.6%Qwen3-VL: Sharper Vision, Deeper Thought, Broader ActionChecked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.thinking modelSource image

All benchmarks