UnifyBench

RankingHumanity's Last Exam

Humanity's Last Exam

Data updated 12 Sept 2026

Bucket
Hard reasoning
Unit
percent
Direction
Higher is better
Version
full
Display harness
HLE no tools
Board
https://lastexam.ai/

Compare published benchmark results with category weights →

Models

ModelScoreEvidenceSource-recorded date
Claude Fable 5.1Anthropic60.9%lab self-report

Table8.1.A and section8.12.1: full2500-question HLE, reasoning-only no-tools, adaptive/auto thinking, maximum effort,1M token cap, no compaction, Opus4.6 grader; trials averaged. Production safeguards may route flagged Fable tasks to Opus models; this is disclosed deployed-model behavior, not an unsafeguarded base-model score.

2026-09-01
Claude Fable 5Anthropic57.8%lab self-report

Table8.1.A and section8.12.1: full2500-question HLE, reasoning-only no-tools, adaptive/auto thinking, maximum effort,1M token cap, no compaction, Opus4.6 grader; trials averaged. Production safeguards may route flagged Fable tasks to Opus models; this is disclosed deployed-model behavior, not an unsafeguarded base-model score.

2026-09-01
Claude Opus 5Anthropic56.6%lab self-report

Table8.1.A and section8.12.1: full2500-question HLE, reasoning-only no-tools, adaptive/auto thinking, maximum effort,1M token cap, no compaction, Opus4.6 grader; trials averaged.

2026-09-01
Gemini 3.1 ProGoogle46.44%official board2026-04-10
Gemini 3.8 FlashGoogle44.52%official board2026-09-09
Qwen 3.8-MaxQwen43.6%lab self-report

Cited Qwen page returned no readable content in this audit; exact model/version, value and protocol could not be independently verified from that source.

2026-08-03
Kimi K3Moonshot43.5%lab self-report

No-tools label alone does not verify full2500-task subset, evaluation configuration and exact observation against the preferred unit.

2026-07-16
DeepSeek V4-Pro 0813DeepSeek42.7%lab self-report

No-tools label alone does not verify full2500-task subset, evaluation configuration and exact observation against the preferred unit.

2026-08-13
Claude Opus 4.7Anthropic36.2%official board2026-04-22
NVIDIA Nemotron 3 UltraNVIDIA26.7%lab self-report

Fixture value corresponds to source BF16 column on the NVFP4 repository page; preserve checkpoint/precision and per-benchmark NeMo configuration before mapping to a preferred unit.

2026-06-04
Claude Opus 4.5Anthropic14.16%official board2025-11-26
Claude Sonnet 4.5Anthropic7.52%official board2025-10-02
Llama 4 MaverickMeta5.68%official board2025-04-10
GPT-4.1OpenAI5.4%official board2025-04-14
Mistral Medium 3Mistral4.52%official board2025-05-13
Agnes 2.5 Pro AlphaAgnes AI———
Amazon Nova 2 LiteAmazon———
Amazon Nova 2 Pro PreviewAmazon———
Amazon Nova LiteAmazon———
Amazon Nova MicroAmazon———
Amazon Nova PremierAmazon———
Amazon Nova ProAmazon———
AutoGLM-Phone-9BZ.ai———
AutoGLM-Phone-9B-MultilingualZ.ai———
aya-101Cohere———
aya-23-35BCohere———
aya-23-8BCohere———
aya-expanse-8bCohere———
aya-vision-8bCohere———
C4Ai Aya Expanse 32BCohere———
C4Ai Aya Vision 32BCohere———
c4ai-command-r7b-arabic-02-2025Cohere———
chatglm-6bZ.ai———
chatglm2-6bZ.ai———
chatglm2-6b-32kZ.ai———
chatglm3-6bZ.ai———
chatglm3-6b-128kZ.ai———
chatglm3-6b-32kZ.ai———
Claude Haiku 4.5Anthropic———
Claude Opus 4.6Anthropic———
Claude Opus 4.8Anthropic———
Claude Sonnet 4.6Anthropic———
Claude Sonnet 5Anthropic———
codegeex2-6bZ.ai———
codegeex4-all-9bZ.ai———
CodeGemma 7B ITGoogle———
Codestral 2508Mistral———
cogagent-9b-20241220Z.ai———
cogagent-chat-hfZ.ai———
cogagent-vqa-hfZ.ai———
cogvlm-chat-hfZ.ai———
cogvlm2-llama3-chat-19BZ.ai———
cogvlm2-llama3-chinese-chat-19BZ.ai———
cogvlm2-video-llama3-chatZ.ai———
CommandCohere———
Command A 03 2025Cohere———
Command A Plus 05 2026Cohere———
Command A Reasoning 08 2025Cohere———
Command A Translate 08 2025Cohere———
Command A Vision 07 2025Cohere———
Command LightCohere———
Command R 03 2024Cohere———
Command R 08 2024Cohere———
Command R Plus 04 2024Cohere———
Command R Plus 08 2024Cohere———
Command R7B 12 2024Cohere———
DeepSeek V4 Flash 0424DeepSeek———
DeepSeek V4-Pro 0424DeepSeek———
DeepSeek-R1-0528DeepSeek———
DeepSeek-R1-0528-Qwen3-8BDeepSeek———
DeepSeek-R1-Distill-Llama-70BDeepSeek———
DeepSeek-R1-Distill-Llama-8BDeepSeek———
DeepSeek-R1-Distill-Qwen-1.5BDeepSeek———
DeepSeek-R1-Distill-Qwen-14BDeepSeek———
DeepSeek-R1-Distill-Qwen-32BDeepSeek———
DeepSeek-R1-Distill-Qwen-7BDeepSeek———
DeepSeek-V3.2DeepSeek———
DeepSeek-V3.2-SpecialeDeepSeek———
DeepSeek-V4-Flash-0731DeepSeek———
DeepSeek-V4-Flash-Vision-ExpDeepSeek———
Devstral 2Mistral———
Devstral Medium 1.0Mistral———
Devstral Small 1.1Mistral———
Devstral Small 2Mistral———
DiffusionGemma 26B A4BGoogle———
Doubao Seed CodeByteDance———
ERNIE-4.5-0.3B-PTBaidu———
ERNIE-4.5-21B-A3B-PTBaidu———
ERNIE-4.5-21B-A3B-ThinkingBaidu———
ERNIE-4.5-300B-A47B-PTBaidu———
ERNIE-4.5-turbo-128kBaidu———
ERNIE-4.5-turbo-20260402Baidu———
ERNIE-4.5-turbo-32kBaidu———
ERNIE-4.5-turbo-vlBaidu———
ERNIE-4.5-turbo-vl-32kBaidu———
ERNIE-4.5-VL-28B-A3B-PTBaidu———
ERNIE-4.5-VL-28B-A3B-ThinkingBaidu———
ERNIE-4.5-VL-424B-A47B-PTBaidu———
ERNIE-5.0Baidu———
ERNIE-5.0-thinking-expBaidu———
ERNIE-5.0-thinking-latestBaidu———
ERNIE-5.0-thinking-previewBaidu———
ERNIE-5.1Baidu———
EXAONE 3.0 7.8B InstructLG AI Research———
EXAONE 3.5 2.4B InstructLG AI Research———
EXAONE 3.5 32B InstructLG AI Research———
EXAONE 3.5 7.8B InstructLG AI Research———
EXAONE 4.0 1.2BLG AI Research———
EXAONE 4.0 32BLG AI Research———
EXAONE 4.0.1 32BLG AI Research———
EXAONE 4.5 33BLG AI Research———
EXAONE Deep 2.4BLG AI Research———
EXAONE Deep 32BLG AI Research———
EXAONE Deep 7.8BLG AI Research———
Gemini 2.5 FlashGoogle———
Gemini 2.5 Flash-LiteGoogle———
Gemini 2.5 ProGoogle———
Gemini 3 Flash PreviewGoogle———
Gemini 3.1 Flash-LiteGoogle———
Gemini 3.5 FlashGoogle———
Gemini 3.5 Flash-LiteGoogle———
Gemini 3.6 FlashGoogle———
Gemini 3.7 FlashGoogle———
Gemma 1 2B ITGoogle———
Gemma 1 7B ITGoogle———
Gemma 1.1 2B ITGoogle———
Gemma 1.1 7B ITGoogle———
Gemma 2 27B ITGoogle———
Gemma 2 2B ITGoogle———
Gemma 2 9B ITGoogle———
Gemma 3 12B ITGoogle———
Gemma 3 1B ITGoogle———
Gemma 3 270M ITGoogle———
Gemma 3 27B ITGoogle———
Gemma 3 4B ITGoogle———
Gemma 3n E2B ITGoogle———
Gemma 3n E4B ITGoogle———
Gemma 4 12B ITGoogle———
Gemma 4 26B-A4B ITGoogle———
Gemma 4 31B ITGoogle———
Gemma 4 E2B ITGoogle———
Gemma 4 E4B ITGoogle———
GLM-4-32B-0414Z.ai———
GLM-4-32B-0414-128KZ.ai———
GLM-4-9B-0414Z.ai———
glm-4-9b-chatZ.ai———
glm-4-9b-chat-1mZ.ai———
GLM-4.1V-9B-ThinkingZ.ai———
GLM-4.5Z.ai———
GLM-4.5-AirZ.ai———
GLM-4.5-AirXZ.ai———
GLM-4.5-FlashZ.ai———
GLM-4.5-XZ.ai———
GLM-4.5VZ.ai———
GLM-4.6Z.ai———
GLM-4.6VZ.ai———
GLM-4.6V-FlashZ.ai———
GLM-4.6V-FlashXZ.ai———
GLM-4.7Z.ai———
GLM-4.7-FlashZ.ai———
GLM-4.7-FlashXZ.ai———
glm-4v-9bZ.ai———
GLM-5Z.ai———
GLM-5-TurboZ.ai———
GLM-5.1Z.ai———
GLM-5.2Z.ai———
GLM-5.3Z.ai———
GLM-5.3-FlashZ.ai———
GLM-5V-TurboZ.ai———
glm-edge-1.5b-chatZ.ai———
glm-edge-4b-chatZ.ai———
glm-edge-v-2bZ.ai———
glm-edge-v-5bZ.ai———
GLM-Z1-32B-0414Z.ai———
GLM-Z1-9B-0414Z.ai———
GLM-Z1-Rumination-32B-0414Z.ai———
GPT-3.5-turboOpenAI———
GPT-3.5-turbo-1106OpenAI———
GPT-3.5-turbo-instructOpenAI———
GPT-4OpenAI———
GPT-4-turboOpenAI———
GPT-4.1 MiniOpenAI———
GPT-4.1-nanoOpenAI———
GPT-4oOpenAI———
GPT-4o MiniOpenAI———
GPT-5OpenAI———
GPT-5 MiniOpenAI———
GPT-5 NanoOpenAI———
GPT-5 ProOpenAI———
GPT-5.1OpenAI———
GPT-5.2OpenAI———
GPT-5.2 ProOpenAI———
GPT-5.3 CodexOpenAI———
GPT-5.3 Codex SparkOpenAI———
GPT-5.4OpenAI———
GPT-5.4 MiniOpenAI———
GPT-5.4 NanoOpenAI———
GPT-5.4 ProOpenAI———
GPT-5.5OpenAI———
GPT-5.5 ProOpenAI———
GPT-5.6 LunaOpenAI———
GPT-5.6 SolOpenAI———
GPT-5.6 TerraOpenAI———
GPT-6 AstraOpenAI———
gpt-oss-120bOpenAI———
gpt-oss-20bOpenAI———
Grok 4.6xAI———
Grok-4.20-0309-non-reasoningxAI———
Grok-4.20-0309-reasoningxAI———
Grok-4.20-multi-agent-0309xAI———
Grok-4.3xAI———
Grok-4.5xAI———
Grok-build-0.1xAI———
Hunyuan A52B InstructTencent———
Hunyuan-0.5B-InstructTencent———
Hunyuan-1.8B-InstructTencent———
Hunyuan-4B-InstructTencent———
Hunyuan-7B-InstructTencent———
Hunyuan-7B-Instruct-0124Tencent———
Hunyuan-A13B-InstructTencent———
Hy3Tencent———
Hy3-previewTencent———
Hy4-previewTencent———
K EXAONE 2.0 750B A37BLG AI Research———
K EXAONE 236B A23BLG AI Research———
Kimi-Dev-72BMoonshot———
Kimi-K2.6Moonshot———
Kimi-K2.7-CodeMoonshot———
Kimi-Linear-48B-A3B-InstructMoonshot———
Kimi-VL-A3B-InstructMoonshot———
Kimi-VL-A3B-Thinking-2506Moonshot———
Leanstral 1.5Mistral———
Llama-3.1-405B-InstructMeta———
Llama-3.1-70B-InstructMeta———
Llama-3.1-8B-InstructMeta———
Llama-3.2-11B-Vision-InstructMeta———
Llama-3.2-1B-InstructMeta———
Llama-3.2-3B-InstructMeta———
Llama-3.2-90B-Vision-InstructMeta———
Llama-3.3-70B-InstructMeta———
Llama-4-Scout-17B-16E-InstructMeta———
Magistral Medium 1.2Mistral———
Magistral Small 1.2Mistral———
MAI-Code-1-FlashMicrosoft———
MAI-Code-1.1-FlashMicrosoft———
MAI-DS-R1Microsoft———
MAI-Thinking-1Microsoft———
MiMo-7B-RLXiaomi———
MiMo-7B-RL-0530Xiaomi———
MiMo-7B-RL-ZeroXiaomi———
MiMo-7B-SFTXiaomi———
MiMo-V2-FlashXiaomi———
MiMo-V2.5Xiaomi———
MiMo-V2.5-ProXiaomi———
MiMo-VL-7B-RLXiaomi———
MiMo-VL-7B-RL-2508Xiaomi———
MiMo-VL-7B-SFTXiaomi———
MiMo-VL-7B-SFT-2508Xiaomi———
MiniCPM5-2BOpenBMB———
MiniMax-M1-40kMiniMax———
MiniMax-M1-80kMiniMax———
MiniMax-M2MiniMax———
MiniMax-M2-herMiniMax———
MiniMax-M2.1MiniMax———
MiniMax-M2.5MiniMax———
MiniMax-M2.7MiniMax———
MiniMax-M3MiniMax———
MiniMax-Text-01MiniMax———
MiniMax-VL-01MiniMax———
Ministral 3 14BMistral———
Ministral 3 3BMistral———
Ministral 3 8BMistral———
Ministral 3BMistral———
Ministral 8BMistral———
Mistral Large 2 (July 2024)Mistral———
Mistral Large 2.1Mistral———
Mistral Large 3Mistral———
Mistral Medium 3.1Mistral———
Mistral Medium 3.5Mistral———
Mistral Nemo 12BMistral———
Mistral Small 3.2Mistral———
Mistral Small 4Mistral———
Mistral Small CreativeMistral———
Muse Glimmer 30BMeta———
Muse Spark 1.1Meta———
Muse Spark 1.2Meta———
Muse Spark 1.3Meta———
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16NVIDIA———
North-Micro-Vision-InstructCohere———
North-Mini-Code-1.0Cohere———
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16NVIDIA———
NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA———
NVIDIA-Nemotron-3-Super-120B-A12B-BF16NVIDIA———
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16NVIDIA———
NVIDIA-Nemotron-Nano-12B-v2NVIDIA———
NVIDIA-Nemotron-Nano-12B-v2-VL-BF16NVIDIA———
NVIDIA-Nemotron-Nano-9B-v2NVIDIA———
NVIDIA-Nemotron-Nano-9B-v2-JapaneseNVIDIA———
o1OpenAI———
o1-proOpenAI———
o3OpenAI———
o3-miniOpenAI———
o3-proOpenAI———
o4-miniOpenAI———
Phi-3-medium-128k-instructMicrosoft———
Phi-3-medium-4k-instructMicrosoft———
Phi-3-mini-128k-instructMicrosoft———
Phi-3-mini-4k-instructMicrosoft———
Phi-3-small-128k-instructMicrosoft———
Phi-3-small-8k-instructMicrosoft———
Phi-3-vision-128k-instructMicrosoft———
Phi-3.5-mini-instructMicrosoft———
Phi-3.5-MoE-instructMicrosoft———
Phi-3.5-vision-instructMicrosoft———
phi-4Microsoft———
Phi-4-mini-flash-reasoningMicrosoft———
Phi-4-mini-instructMicrosoft———
Phi-4-mini-reasoningMicrosoft———
Phi-4-multimodal-instructMicrosoft———
Phi-4-reasoningMicrosoft———
Phi-4-reasoning-plusMicrosoft———
Phi-4-reasoning-vision-15BMicrosoft———
Phi-mini-MoE-instructMicrosoft———
Phi-tiny-MoE-instructMicrosoft———
Pixtral 12BMistral———
Pixtral LargeMistral———
Qianfan-VL-3BBaidu———
Qianfan-VL-70BBaidu———
Qianfan-VL-8BBaidu———
QVQ-maxQwen———
QVQ-plusQwen———
Qwen-coder-plusQwen———
Qwen-coder-turboQwen———
Qwen-flashQwen———
Qwen-flash-characterQwen———
Qwen-longQwen———
Qwen-math-plusQwen———
Qwen-math-turboQwen———
Qwen-maxQwen———
Qwen-omni-turboQwen———
Qwen-plusQwen———
Qwen-plus-characterQwen———
Qwen-plus-character-jaQwen———
Qwen-turboQwen———
Qwen-vl-maxQwen———
Qwen-vl-plusQwen———
Qwen3-0.6BQwen———
Qwen3-1.7BQwen———
Qwen3-14BQwen———
Qwen3-235B-A22BQwen———
Qwen3-235B-A22B-Instruct-2507Qwen———
Qwen3-235B-A22B-Thinking-2507Qwen———
Qwen3-30B-A3BQwen———
Qwen3-30B-A3B-Instruct-2507Qwen———
Qwen3-30B-A3B-Thinking-2507Qwen———
Qwen3-32BQwen———
Qwen3-4BQwen———
Qwen3-4B-Instruct-2507Qwen———
Qwen3-4B-Thinking-2507Qwen———
Qwen3-8BQwen———
Qwen3-Coder-30B-A3B-InstructQwen———
Qwen3-Coder-480B-A35B-InstructQwen———
Qwen3-coder-flashQwen———
Qwen3-Coder-NextQwen———
Qwen3-coder-plusQwen———
Qwen3-maxQwen———
Qwen3-Next-80B-A3B-InstructQwen———
Qwen3-Next-80B-A3B-ThinkingQwen———
Qwen3-Omni-30B-A3B-InstructQwen———
Qwen3-Omni-30B-A3B-ThinkingQwen———
Qwen3-omni-flashQwen———
Qwen3-VL-235B-A22B-InstructQwen———
Qwen3-VL-235B-A22B-ThinkingQwen———
Qwen3-VL-2B-InstructQwen———
Qwen3-VL-2B-ThinkingQwen———
Qwen3-VL-30B-A3B-InstructQwen———
Qwen3-VL-30B-A3B-ThinkingQwen———
Qwen3-VL-32B-InstructQwen———
Qwen3-VL-32B-ThinkingQwen———
Qwen3-VL-4B-InstructQwen———
Qwen3-VL-4B-ThinkingQwen———
Qwen3-VL-8B-InstructQwen———
Qwen3-VL-8B-ThinkingQwen———
Qwen3-vl-flashQwen———
Qwen3-vl-plusQwen———
Qwen3.5-0.8BQwen———
Qwen3.5-122B-A10BQwen———
Qwen3.5-27BQwen———
Qwen3.5-2BQwen———
Qwen3.5-35B-A3BQwen———
Qwen3.5-397B-A17BQwen———
Qwen3.5-4BQwen———
Qwen3.5-9BQwen———
Qwen3.5-flashQwen———
Qwen3.5-omni-flashQwen———
Qwen3.5-omni-plusQwen———
Qwen3.5-plusQwen———
Qwen3.6-27BQwen———
Qwen3.6-35B-A3BQwen———
Qwen3.6-flashQwen———
Qwen3.6-plusQwen———
Qwen3.7-flashQwen———
Qwen3.7-maxQwen———
Qwen3.7-plusQwen———
Qwen3.8-2.4T-A95BQwen———
Qwen3.8-27BQwen———
Qwen3.8-flashQwen———
Qwen3.8-Flash-NextQwen———
QwQ-plusQwen———
RecurrentGemma 2B ITGoogle———
RecurrentGemma 9B ITGoogle———
Seed 1.6ByteDance———
Seed 1.6 ThinkingByteDance———
Seed 1.6 VisionByteDance———
Seed 1.8ByteDance———
Seed 2.0 CodeByteDance———
Seed 2.0 LiteByteDance———
Seed 2.0 MiniByteDance———
Seed 2.0 ProByteDance———
Seed 2.1 ProByteDance———
Seed 2.1 TurboByteDance———
Seed CharacterByteDance———
Seed Coder 8B InstructByteDance———
Seed Coder 8B ReasoningByteDance———
Seed EvolvingByteDance———
Seed OSS 36B InstructByteDance———
SOLAR 10.7B Instruct v1.0Upstage———
Solar Mini (April 2025)Upstage———
Solar Open 100BUpstage———
Solar Open2 250BUpstage———
Solar Pro 2 (December 2025)Upstage———
Solar Pro 3Upstage———
Solar Pro 4Upstage———
solar pro preview instructUpstage———
Step 3StepFun———
Step 3 VL 10BStepFun———
Step 3.5 FlashStepFun———
Step 3.7 FlashStepFun———
SynLogic-32BMiniMax———
SynLogic-7BMiniMax———
SynLogic-Mix-3-32BMiniMax———
Tiny Aya EarthCohere———
Tiny Aya FireCohere———
Tiny Aya GlobalCohere———
Tiny Aya WaterCohere———
visualglm-6bZ.ai———
Voxtral MiniMistral———
Voxtral SmallMistral———
WeDLM-7B-InstructTencent———
WeDLM-8B-InstructTencent———
Youtu-LLM-2BTencent———

Other harnesses

These runs use other harnesses. Capability scoring matches configurations separately; model details identify which source records contribute.

ModelScoreHarnessEvidenceSource-recorded date
Claude Fable 5Anthropic63.8%HLE with toolslab self-report

Same full HLE with web search, restricted web fetch, programmatic tools and code execution. Kept separate from no-tools model capability; production safeguard fallback applies.

2026-09-01
Claude Fable 5.1Anthropic65%HLE with toolslab self-report

Same full HLE with web search, restricted web fetch, programmatic tools and code execution. Kept separate from no-tools model capability; production safeguard fallback applies.

2026-09-01
Claude Opus 5Anthropic63.6%HLE with toolslab self-report

Same full HLE with web search, restricted web fetch, programmatic tools and code execution. Kept separate from no-tools model capability.

2026-09-01
Claude Opus 5Anthropic64.7%HLE with toolslab self-report

Tool-enabled result is explicitly incompatible with the preferred no-tools unit.

2026-07-24