Best LLMs for advanced reasoning

Humanity’s Last Exam

Humanity’s Last Exam evaluates advanced knowledge and reasoning across a broad range of academic subjects using expert-written questions. It includes text and multimodal problems designed to challenge frontier AI systems. Results identify whether the full exam or a text-only subset was used.

Updated: Nov 5, 2026Data as of: Nov 2, 2026

Widely usedNot saturated

Official benchmark

Humanity’s Last Exam results. Each row identifies its benchmark version and evaluation track. Scores from different versions are not directly comparable. Best effort from the latest stored report date per model and version.
Model / creator	Version	Reported score
Kimi K3.1 ProMoonshot AI · Open source / open weights	Full set	70%
Claude Fable 5.1Anthropic · Proprietary	Full set	65%
Claude Mythos PreviewAnthropic · Proprietary	Full set	64.7%
Claude Mythos 5Anthropic · Proprietary	Full set	64.5%
GPT-5.6 SolOpenAI · Proprietary	Unspecified subset	64.5%
Claude Opus 5.5Anthropic · Proprietary	Full set	64.4%
DeepSeek V4.1 FlashDeepSeek · Open source / open weights	Unspecified subset	63.9%
Claude Opus 5Anthropic · Proprietary	Full set	63.6%
Claude Fable 5Anthropic · Proprietary	Full set	63%
GLM-5.3Z.ai · Open source / open weights	Unspecified subset	62.5%
Muse Spark 1.1Meta · Proprietary	Full set	62.1%
Claude Opus 5.5Anthropic · Proprietary	Text only	61.4%
DeepSeek V4 Pro 0813DeepSeek · Open source / open weights	Unspecified subset	60%
Kimi K3Moonshot AI · Open source / open weights	Unspecified subset	59.8%
Claude Fable 5.1Anthropic · Proprietary	Text only	59.1%
DeepSeek V4.1 Flash 1017DeepSeek · Open source / open weights	Full set	58.8%
GPT-5.4 ProOpenAI · Proprietary	Full set	58.7%
GPT-5.6 SolOpenAI · Proprietary	Full set	58%
Muse SparkMeta · Proprietary	Unspecified subset	58%
Claude Opus 4.8Anthropic · Proprietary	Full set	57.9%
Claude Haiku 5.5Anthropic · Proprietary	Full set	57.4%
Claude Sonnet 5Anthropic · Proprietary	Full set	57.4%
GPT-5.5 ProOpenAI · Proprietary	Full set	57.2%
GPT-6 AstraOpenAI · Proprietary	Full set	57.2%
Gemini 4 ArgonGoogle · Proprietary	Text only	57.1%
Claude Sonnet 5.5Anthropic · Proprietary	Full set	56.9%
Kimi K3Moonshot AI · Open source / open weights	Full set	56%
Claude Fable 5Anthropic · Proprietary	Text only	55.5%
GLM-5.3-FlashZ.ai · Open source / open weights	Full set	55.3%
Claude Sonnet 5.5Anthropic · Proprietary	Text only	55%
Claude Opus 5Anthropic · Proprietary	Text only	54.9%
Claude Opus 4.7Anthropic · Proprietary	Full set	54.7%
GLM-5.2Z.ai · Open source / open weights	Text only	54.7%
GPT-6 AstraOpenAI · Proprietary	Text only	54.68%
Kimi K2.6Moonshot AI · Open source / open weights	Unspecified subset	54%
Claude Opus 4.6Anthropic · Proprietary	Full set	53.1%
GPT-6.1 SolOpenAI · Proprietary	Text only	52.919%
Kimi K3.1 ProMoonshot AI · Open source / open weights	Text only	52.5%
GPT-5.5OpenAI · Proprietary	Full set	52.2%
GPT-5.4OpenAI · Proprietary	Full set	52.1%
Gemini 3.1 ProGoogle · Proprietary	Full set	51.4%
Kimi K2 ThinkingMoonshot AI · Open source / open weights	Text only	51%
Grok 4 HeavyxAI · Proprietary	Text only	50.7%
Muse SparkMeta · Proprietary	Full set	50.4%
Kimi K2.5Moonshot AI · Open source / open weights	Full set	50.2%
Muse Wonder 1.0Meta · Proprietary	Text only	50.1%
GPT-5.6 SolOpenAI · Proprietary	Text only	49.5%
MiMo-V2.6-ProXiaomi · Open source / open weights	Text only	49.4%
Claude Opus 4.8Anthropic · Proprietary	Text only	48.7%
Muse Spark 1.3Meta · Proprietary	Text only	48.7%
DeepSeek V4 Pro 0424DeepSeek · Open source / open weights	Unspecified subset	48.2%
MiniMax M3.1 FlashMiniMax · Open source / open weights	Text only	48%
Gemini 3.8 FlashGoogle · Proprietary	Text only	47.8%
Gemini 3.1 ProGoogle · Proprietary	Text only	47%
Kimi K3Moonshot AI · Open source / open weights	Text only	46.9%
Claude Sonnet 4.6Anthropic · Proprietary	Full set	46.8%
Step 5 PreviewStepFun · Open source / open weights	Text only	46.5%
GPT-6.1 TerraOpenAI · Proprietary	Text only	46.2%
Muse Spark 1.1Meta · Proprietary	Text only	46.2%
Gemini 3 ProGoogle · Proprietary	Full set	45.8%
GPT-5.5OpenAI · Proprietary	Text only	45.8%
GPT-5.2OpenAI · Proprietary	Full set	45.5%
Muse Spark 1.2Meta · Proprietary	Text only	45.5%
DeepSeek V4 Flash 0420DeepSeek · Open source / open weights	Unspecified subset	45.1%
Claude Haiku 5.5Anthropic · Proprietary	Text only	44.4%
Muse Spark OpenMeta · Open source / open weights	Text only	43.9%
GPT-5.4OpenAI · Proprietary	Text only	43.7%
DeepSeek V4.1 Flash 1017DeepSeek · Open source / open weights	Text only	43.3%
Claude Opus 4.5Anthropic · Proprietary	Full set	43.2%
Grok 4.7xAI · Proprietary	Text only	43.1%
GPT-5.6 TerraOpenAI · Proprietary	Text only	42.9%
Claude Opus 4.7Anthropic · Proprietary	Text only	42.3%
GLM-5.3Z.ai · Open source / open weights	Text only	42.3%
Claude Sonnet 5Anthropic · Proprietary	Text only	41.3%
DeepSeek V4 Pro 0813DeepSeek · Open source / open weights	Text only	41%
DeepSeek V3.2DeepSeek · Open source / open weights	Text only	40.8%
Muse SparkMeta · Proprietary	Text only	40.7%
Claude Opus 4.6Anthropic · Proprietary	Text only	39.9%
GLM-5.3-FlashZ.ai · Open source / open weights	Text only	39.852%
Gemini 3 ProGoogle · Proprietary	Text only	39.7%
GPT-5.6 LunaOpenAI · Proprietary	Text only	39.5%
DeepSeek V4.1 FlashDeepSeek · Open source / open weights	Text only	39.2%
MiniMax M3MiniMax · Open source / open weights	Text only	38.971%
GPT-6 LunaOpenAI · Proprietary	Text only	38.5%
GPT-5.2OpenAI · Proprietary	Text only	37.7%
DeepSeek V4 Pro 0424DeepSeek · Open source / open weights	Text only	37.535%
Kimi K2.6Moonshot AI · Open source / open weights	Text only	37.488%
DeepSeek V4.1 FlashDeepSeek · Open source / open weights	Full set	36.8%
Gemini 3 FlashGoogle · Proprietary	Text only	36.6%
DeepSeek V4 Flash 0420DeepSeek · Open source / open weights	Text only	34.847%
Qwen3.8 27BQwen · Open source / open weights	Text only	33.9%
Claude Sonnet 4.6Anthropic · Proprietary	Text only	33.6%
InklingThinking Machines · Open source / open weights	Text only	31.9%
Kimi K2.5Moonshot AI · Open source / open weights	Text only	30.723%
Nemotron 3.5 Super VLNVIDIA · Open source / open weights	Text only	30.7%
Claude Opus 4.5Anthropic · Proprietary	Text only	30.1%
GLM-5.1Z.ai · Open source / open weights	Text only	30.1%
MiniMax M2.7MiniMax · Open source / open weights	Text only	29.6%
GLM-5Z.ai · Open source / open weights	Text only	29.3%
Qwen3.5-397B-A17BQwen · Open source / open weights	Text only	29%
GPT-5.4 miniOpenAI · Proprietary	Text only	28.1%
Qwen3.6 PlusQwen · Proprietary	Text only	27.8%
Step 3.7 FlashStepFun · Open source / open weights	Text only	21.4%
Qwen 3.5 35B-A3BQwen · Open source / open weights	Text only	21%
gpt-oss-120bOpenAI · Open source / open weights	Text only	19.601%
Claude Sonnet 4.5Anthropic · Proprietary	Text only	17.8%
Qwen 3.5 9BQwen · Open source / open weights	Text only	14.9%
gpt-oss-20bOpenAI · Open source / open weights	Text only	10.982%
DeepSeek R1 (Jan 2025)DeepSeek · Open source / open weights	Text only	8.5%
GPT-4.1 miniOpenAI · Proprietary	Text only	5%
Qwen3-14BQwen · Open source / open weights	Text only	4.5%
GPT-4.1OpenAI · Proprietary	Text only	4.2%
GPT-4 TurboOpenAI · Proprietary	Text only	3.1%

View all 173 published reports

Claude Mythos 564.5%

Full set

Full 2,500 questions; thinking auto; total 1M-token cap; no context compaction; Opus 4.6 judge. Tools variant: search, fetch, programmatic tool calling and code execution. · subset: full · Tools enabled

Anthropic ↗

Claude Opus 5.564.4%

Full set

Full 2,500 questions; no tools; thinking auto; total 1M-token cap. Source-specific production fallback configuration. · subset: full · No tools

Anthropic ↗

Claude Fable 563%

Full set

Full-set comparison; reasoning effort and tool configuration as reported. Kept separate from later publications. · subset: full · Tools enabled

Moonshot AI ↗

Claude Mythos 559%

Full set

Full 2,500 questions; thinking auto; total 1M-token cap; no context compaction; Opus 4.6 judge. Tools variant: search, fetch, programmatic tool calling and code execution. · subset: full · No tools

Anthropic ↗

GPT-5.6 Sol58%

Full set

Full-set comparison; reasoning effort and tool configuration as reported. Kept separate from later publications. · subset: full · Tools enabled

Moonshot AI ↗

Claude Sonnet 557.4%

Full set

Anthropic ↗

Claude Fable 553.3%

Full set

Full-set comparison; reasoning effort and tool configuration as reported. Kept separate from later publications. · subset: full · No tools

Moonshot AI ↗

Claude Opus 4.653.1%

Full set

Opus 4.6 comparison in Mythos Preview launch table. · subset: full · Tools enabled

Anthropic ↗

Claude Sonnet 4.646.8%

Full set

Sonnet 4.6 comparison; full 2,500-question HLE; no-tools versus tools. · subset: full · Tools enabled

Anthropic ↗

Gemini 3 Pro45.8%

Full set

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: full · Tools enabled

Moonshot AI ↗

GPT-5.245.5%

Full set

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: full · Tools enabled

Moonshot AI ↗

GPT-5.6 Sol44.5%

Full set

Full-set comparison; reasoning effort and tool configuration as reported. Kept separate from later publications. · subset: full · No tools

Moonshot AI ↗

Claude Opus 4.543.2%

Full set

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: full · Tools enabled

Moonshot AI ↗

Claude Sonnet 543.2%

Full set

Full 2,500 questions; thinking auto; total 1M-token cap; no context compaction; Opus 4.6 judge. Tools variant: search, fetch, programmatic tool calling and code execution. · subset: full · No tools

Anthropic ↗

Claude Opus 4.640%

Full set

Opus 4.6 comparison in Mythos Preview launch table. · subset: full · No tools

Anthropic ↗

Gemini 3 Pro37.5%

Full set

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: full · No tools

Moonshot AI ↗

Claude Sonnet 4.634.6%

Full set

Sonnet 4.6 comparison; full 2,500-question HLE; no-tools versus tools. · subset: full · No tools

Anthropic ↗

GPT-5.234.5%

Full set

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: full · No tools

Moonshot AI ↗

Claude Opus 4.530.8%

Full set

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: full · No tools

Moonshot AI ↗

Kimi K3.1 Pro70%

Full set

Max effort; temperature=1.0; top_p=0.95; full 2500-question set. · subset: full · Tools enabled

Moonshot AI ↗

Claude Fable 5.165%

Full set

subset: full · Tools enabled

Anthropic ↗

Claude Mythos Preview64.7%

Full set

subset: full · Tools enabled

Anthropic ↗

Claude Fable 563.8%

Full set

subset: full · Tools enabled

Anthropic ↗

Claude Opus 563.6%

Full set

subset: full · Tools enabled

Anthropic ↗

Muse Spark 1.162.1%

Full set

Full 2500 questions; o3-mini judge; bash and browser tools; source describes domain blocklist. · subset: full · Tools enabled

Meta ↗

Claude Fable 5.160.9%

Full set

subset: full · No tools

Anthropic ↗

DeepSeek V4.1 Flash 101758.8%

Full set

Max effort; competitor result in Moonshot AI report. · subset: full · Tools enabled

Moonshot AI ↗

Claude Fable 557.8%

Full set

subset: full · No tools

Anthropic ↗

GPT-6 Astra57.2%

Full set

OpenAI Astra launch comparison; reported with tools. Full-set interpretation corroborated by Moonshot’s full-set comparison. · subset: full · Tools enabled

OpenAI ↗

GPT-6 Astra57.2%

Full set

Max effort as labeled in Moonshot comparison; GPT models may include cyberguards. Confirms OpenAI launch table HLE with tools 57.2. · subset: full · Tools enabled

Moonshot AI ↗

Claude Mythos Preview56.8%

Full set

subset: full · No tools

Anthropic ↗

Claude Opus 556.6%

Full set

subset: full · No tools

Anthropic ↗

Kimi K356%

Full set

Max effort; temperature=1.0; top_p=1.0; general tools. · subset: full · Tools enabled

Moonshot AI ↗

GLM-5.3-Flash55.3%

Full set

Primary bench_53.png; full set; temperature 1; top_p .95; 163840 generation; 300K context; GPT-5.6 Luna judge · Tools enabled

Z.ai ↗

Muse Spark 1.152.2%

Full set

Full 2500 questions; o3-mini judge. · subset: full · No tools

Meta ↗

Kimi K3.1 Pro51.1%

Full set

Max effort; temperature=1.0; top_p=0.95; full 2500-question set. · subset: full · No tools

Moonshot AI ↗

Muse Spark50.4%

Full set

Later Meta evaluation report; keep separate from launch Contemplating-mode result. · subset: full · Tools enabled

Meta ↗

Kimi K2.550.2%

Full set

Full text and image set; search, code and browsing; 96K completion; simple context management; Hugging Face blocked · Tools enabled

Moonshot AI ↗

Kimi K343.5%

Full set

Max effort; temperature=1.0; top_p=0.95. · subset: full · No tools

Moonshot AI ↗

Muse Spark42.8%

Full set

subset: full · No tools

Meta ↗

Kimi K2.530.1%

Full set

Full text and image set; no tools; 96K completion · No tools

Moonshot AI ↗

DeepSeek V4.1 Flash 101742.6%

Full set

subset: full · No tools

DeepSeek ↗

Claude Haiku 5.557.4%

Full set

Haiku 5.5 launch comparison; reported no-tools and with-tools configurations. · subset: full · Tools enabled

Anthropic ↗

Claude Sonnet 5.556.9%

Full set

Haiku 5.5 launch comparison; reported no-tools and with-tools configurations. · subset: full · No tools

Anthropic ↗

Claude Haiku 5.545.9%

Full set

Haiku 5.5 launch comparison; reported no-tools and with-tools configurations. · subset: full · No tools

Anthropic ↗

Claude Sonnet 5.564.5%

Full set

subset: full · Tools enabled

Anthropic ↗

Claude Sonnet 554.9%

Full set

subset: full · Tools enabled

Anthropic ↗

Claude Opus 5.567.7%

Full set

Provider benchmark table; tools enabled. Cross-model settings may differ. · subset: full · Tools enabled

Anthropic ↗

Claude Fable 5.165.6%

Full set

Provider benchmark table; tools enabled. Cross-model settings may differ. · subset: full · Tools enabled

Anthropic ↗

Claude Opus 563.6%

Full set

Provider benchmark table; tools enabled. Cross-model settings may differ. · subset: full · Tools enabled

Anthropic ↗

DeepSeek V4.1 Flash36.8%

Full set

subset: full · No tools

DeepSeek ↗

Claude Opus 4.857.9%

Full set

subset: full · Tools enabled

Z.ai ↗

Claude Opus 4.849.8%

Full set

subset: full · No tools

Z.ai ↗

GPT-5.4 Pro58.7%

Full set

Reasoning effort xhigh; research environment. · subset: full · Tools enabled

OpenAI ↗

GPT-5.5 Pro57.2%

Full set

Reasoning effort xhigh; research environment. · subset: full · Tools enabled

OpenAI ↗

Claude Opus 4.754.7%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · Tools enabled

OpenAI ↗

GPT-5.552.2%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · Tools enabled

OpenAI ↗

GPT-5.452.1%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · Tools enabled

OpenAI ↗

Gemini 3.1 Pro51.4%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · Tools enabled

OpenAI ↗

Claude Opus 4.746.9%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · No tools

OpenAI ↗

Gemini 3.1 Pro44.4%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · No tools

OpenAI ↗

GPT-5.5 Pro43.1%

Full set

Reasoning effort xhigh; research environment. · subset: full · No tools

OpenAI ↗

GPT-5.4 Pro42.7%

Full set

Reasoning effort xhigh; research environment. · subset: full · No tools

OpenAI ↗

GPT-5.541.4%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · No tools

OpenAI ↗

GPT-5.439.8%

Full set

Launch comparison; GPT models evaluated at xhigh effort; research environment. Competitor settings follow source footnotes. · subset: full · No tools

OpenAI ↗

Claude Opus 5.561.4%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Opus 5.5 (max with fallback); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-opus-5-5 · subset: text · No tools

Claude Fable 5.159.1%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Fable 5.1 (max with fallback); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-fable-5-1 · subset: text · No tools

Claude Fable 555.5%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Fable 5 (with fallback); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-fable-5 · subset: text · No tools

Claude Sonnet 5.555%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Sonnet 5.5 (max with fallback); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-sonnet-5-5 · subset: text · No tools

Claude Opus 554.9%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Opus 5 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-opus-5 · subset: text · No tools

Kimi K3.1 Pro52.5%

Text only

question count: 2158 · reported precision decimals: 1 · Kimi K3.1 Pro; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/kimi-k3-1-pro · subset: text · No tools

Kimi K2 Thinking51%

Text only

Heavy configuration: parallel test-time compute. Text-only; retained separately from single-agent 44.9. · subset: text · Tools enabled

Moonshot AI ↗

Muse Wonder 1.050.1%

Text only

question count: 2158 · reported precision decimals: 1 · Muse Wonder (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/muse-wonder · subset: text · No tools

GPT-5.6 Sol49.5%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-5.6 Sol (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-5-6-sol · subset: text · No tools

MiMo-V2.6-Pro49.4%

Text only

question count: 2158 · reported precision decimals: 1 · MiMo-V2.6-Pro; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/mimo-v2-6-pro · subset: text · No tools

Claude Opus 4.848.7%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Opus 4.8 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-opus-4-8 · subset: text · No tools

Muse Spark 1.348.7%

Text only

question count: 2158 · reported precision decimals: 1 · Muse Spark 1.3 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/muse-spark-1-3 · subset: text · No tools

MiniMax M3.1 Flash48%

Text only

question count: 2158 · reported precision decimals: 1 · MiniMax-M3.1-Flash; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/minimax-m3-1-flash · subset: text · No tools

Gemini 3.8 Flash47.8%

Text only

question count: 2158 · reported precision decimals: 1 · Gemini 3.8 Flash (high); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gemini-3-8-flash · subset: text · No tools

Gemini 3.1 Pro47%

Text only

question count: 2158 · reported precision decimals: 1 · Gemini 3.1 Pro Preview; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gemini-3-1-pro-preview · subset: text · No tools

Kimi K346.9%

Text only

question count: 2158 · reported precision decimals: 1 · Kimi K3 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/kimi-k3 · subset: text · No tools

Step 5 Preview46.5%

Text only

question count: 2158 · reported precision decimals: 1 · Step 5 Preview; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/step-5 · subset: text · No tools

Muse Spark 1.146.2%

Text only

question count: 2158 · reported precision decimals: 1 · Muse Spark 1.1 (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/muse-spark-1-1 · subset: text · No tools

GPT-6.1 Terra46.2%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-6.1 Terra (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-6-1-terra · subset: text · No tools

GPT-5.545.8%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-5.5 (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-5-5 · subset: text · No tools

Muse Spark 1.245.5%

Text only

question count: 2158 · reported precision decimals: 1 · Muse Spark 1.2 (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/muse-spark-1-2 · subset: text · No tools

Claude Haiku 5.544.4%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Haiku 5.5 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-haiku-5-5 · subset: text · No tools

Muse Spark Open43.9%

Text only

question count: 2158 · reported precision decimals: 1 · Muse Spark Open (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/muse-spark-open · subset: text · No tools

GPT-5.443.7%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-5.4 (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-5-4 · subset: text · No tools

DeepSeek V4.1 Flash 101743.3%

Text only

question count: 2158 · reported precision decimals: 1 · DeepSeek V4.1 Flash 1017 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/deepseek-v4-1-flash-1017 · subset: text · No tools

GPT-5.6 Terra42.9%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-5.6 Terra (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-5-6-terra · subset: text · No tools

GLM-5.342.3%

Text only

question count: 2158 · reported precision decimals: 1 · GLM-5.3 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/glm-5-3 · subset: text · No tools

Claude Opus 4.742.3%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Opus 4.7 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-opus-4-7 · subset: text · No tools

Claude Sonnet 541.3%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Sonnet 5 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-sonnet-5 · subset: text · No tools

DeepSeek V4 Pro 081341%

Text only

question count: 2158 · reported precision decimals: 1 · DeepSeek V4 Pro 0813 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/deepseek-v4-pro · subset: text · No tools

DeepSeek V3.240.8%

Text only

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: text · Tools enabled

Moonshot AI ↗

Muse Spark40.7%

Text only

question count: 2158 · reported precision decimals: 1 · Muse Spark; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/muse-spark · subset: text · No tools

Claude Opus 4.639.9%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Opus 4.6 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-opus-4-6-adaptive · subset: text · No tools

Gemini 3 Pro39.7%

Text only

question count: 2158 · reported precision decimals: 1 · Gemini 3 Pro Preview (high); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gemini-3-pro · subset: text · No tools

GPT-5.6 Luna39.5%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-5.6 Luna (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-5-6-luna · subset: text · No tools

DeepSeek V4.1 Flash39.2%

Text only

question count: 2158 · reported precision decimals: 1 · DeepSeek V4.1 Flash (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/deepseek-v4-1-flash · subset: text · No tools

GPT-6 Luna38.5%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-6 Luna (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-6-luna · subset: text · No tools

GPT-5.237.7%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-5.2 (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-5-2 · subset: text · No tools

Gemini 3 Flash36.6%

Text only

question count: 2158 · reported precision decimals: 1 · Gemini 3 Flash; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gemini-3-flash-reasoning · subset: text · No tools

Qwen3.8 27B33.9%

Text only

question count: 2158 · reported precision decimals: 1 · Qwen3.8 27B (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/qwen3-8-27b · subset: text · No tools

Claude Sonnet 4.633.6%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Sonnet 4.6 (max); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-sonnet-4-6-adaptive · subset: text · No tools

Inkling31.9%

Text only

question count: 2158 · reported precision decimals: 1 · Inkling (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/inkling · subset: text · No tools

Nemotron 3.5 Super VL30.7%

Text only

question count: 2158 · reported precision decimals: 1 · Nemotron 3.5 Super VL; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/nemotron-3-5-super-vl · subset: text · No tools

GLM-5.130.1%

Text only

question count: 2158 · reported precision decimals: 1 · GLM-5.1; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/glm-5-1 · subset: text · No tools

Claude Opus 4.530.1%

Text only

question count: 2158 · reported precision decimals: 1 · Claude Opus 4.5; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-opus-4-5-thinking · subset: text · No tools

MiniMax M2.729.6%

Text only

question count: 2158 · reported precision decimals: 1 · MiniMax-M2.7; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/minimax-m2-7 · subset: text · No tools

GLM-529.3%

Text only

question count: 2158 · reported precision decimals: 1 · GLM-5; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/glm-5 · subset: text · No tools

Qwen3.5-397B-A17B29%

Text only

question count: 2158 · reported precision decimals: 1 · Qwen3.5 397B A17B; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/qwen3-5-397b-a17b · subset: text · No tools

GPT-5.4 mini28.1%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-5.4 mini (xhigh); original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-5-4-mini · subset: text · No tools

Qwen3.6 Plus27.8%

Text only

question count: 2158 · reported precision decimals: 1 · Qwen3.6 Plus; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/qwen3-6-plus · subset: text · No tools

DeepSeek V3.225.1%

Text only

Published comparison table; DeepSeek dagger denotes text-only. Other rows use full text and image set. · subset: text · No tools

Moonshot AI ↗

DeepSeek V3.224.6%

Text only

question count: 2158 · reported precision decimals: 1 · DeepSeek V3.2; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/deepseek-v3-2-reasoning · subset: text · No tools

Kimi K2 Thinking23.9%

Text only

Provider no-tools text-only evaluation. Distinct from independent AA 23.8. · subset: text · No tools

Moonshot AI ↗

Step 3.7 Flash21.4%

Text only

question count: 2158 · reported precision decimals: 1 · Step 3.7 Flash; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/step-3-7-flash · subset: text · No tools

Qwen 3.5 35B-A3B21%

Text only

question count: 2158 · reported precision decimals: 1 · Qwen3.5 35B A3B; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/qwen3-5-35b-a3b · subset: text · No tools

Claude Sonnet 4.517.8%

Text only

question count: 2158 · reported precision decimals: 1 · Claude 4.5 Sonnet; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/claude-4-5-sonnet-thinking · subset: text · No tools

Qwen 3.5 9B14.9%

Text only

question count: 2158 · reported precision decimals: 1 · Qwen3.5 9B; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/qwen3-5-9b · subset: text · No tools

GPT-4.1 mini5%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-4.1 mini; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-4-1-mini · subset: text · No tools

Qwen3-14B4.5%

Text only

question count: 2158 · reported precision decimals: 1 · Qwen3 14B; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/qwen3-14b-instruct-reasoning · subset: text · No tools

GPT-4.14.2%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-4.1; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-4-1 · subset: text · No tools

GPT-4 Turbo3.1%

Text only

question count: 2158 · reported precision decimals: 1 · GPT-4 Turbo; original HLE; 2,158 text-only questions; no tools; pass@1; chart labels rounded to one decimal. · source model href: /models/gpt-4-turbo · subset: text · No tools

Gemini 4 Argon57.1%

Text only

High effort; AA methodology: 2158 text-only questions; pass@1; no tools; GPT-5.6 Luna medium judge. · subset: text · No tools

GLM-5.254.7%

Text only

Text-only per model-card footnote; 163840 generation; 300K context; no context management; GPT-5.5 medium judge · Tools enabled

Z.ai ↗

GPT-6 Astra54.68%

Text only

GPT-6 Astra (Max); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6 Astra54.588%

Text only

GPT-6 Astra (Xhigh); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6 Astra53.058%

Text only

GPT-6 Astra (High); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6.1 Sol52.919%

Text only

GPT-6.1 Sol (Max); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6 Astra52.734%

Text only

GPT-6 Astra (Medium); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6.1 Sol52.641%

Text only

GPT-6.1 Sol (Xhigh); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6.1 Sol51.39%

Text only

GPT-6.1 Sol (High); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6.1 Sol49.861%

Text only

GPT-6.1 Sol (Medium); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6 Astra49.212%

Text only

GPT-6 Astra (Low); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

GPT-6.1 Sol47.359%

Text only

GPT-6.1 Sol (Low); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools

Kimi K2 Thinking44.9%

Text only

Text-only; 120 steps; 48K reasoning per step; 256K context; o3-mini judge; Hugging Face blocked · Tools enabled

Moonshot AI ↗

Grok 4.743.1%

Text only

xhigh effort; AA methodology: 2158 text-only questions; pass@1; no tools; GPT-5.6 Luna medium judge. · subset: text · No tools

GLM-5.241.149%

Text only

GLM-5.2 (Max); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

GLM-5.3-Flash39.852%

Text only

GLM-5.3-Flash; original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

MiniMax M338.971%

Text only

MiniMax-M3; original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek V4 Pro 042437.535%

Text only

DeepSeek V4 Pro 0424 (Max); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

Kimi K2.637.488%

Text only

Kimi K2.6 (Reasoning); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek V4 Pro 042435.218%

Text only

DeepSeek V4 Pro 0424 (High); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek V4 Flash 042034.847%

Text only

DeepSeek V4 Flash 0420 (Max); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

Kimi K2.530.723%

Text only

Kimi K2.5 (Reasoning); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek V4 Flash 042030.259%

Text only

DeepSeek V4 Flash 0420 (High); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

Kimi K2 Thinking23.818%

Text only

Kimi K2 Thinking; original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

gpt-oss-120b19.601%

Text only

gpt-oss-120b (High); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

Kimi K2.619.555%

Text only

Kimi K2.6 (Non-reasoning); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

Kimi K2.513.16%

Text only

Kimi K2.5 (Non-reasoning); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

gpt-oss-20b10.982%

Text only

gpt-oss-20b (High); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

GLM-5.29.778%

Text only

GLM-5.2 (Non-reasoning); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek R1 (Jan 2025)8.5%

Text only

DeepSeek R1 (Jan '25); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek V4 Pro 04248.248%

Text only

DeepSeek V4 Pro 0424 (Non-reasoning); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek V4 Flash 04207.785%

Text only

DeepSeek V4 Flash 0420 (Non-reasoning); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

gpt-oss-120b5.885%

Text only

gpt-oss-120b (Low); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

gpt-oss-20b5.283%

Text only

gpt-oss-20b (Low); original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools

DeepSeek V4.1 Flash 101744%

Text only

subset: text · No tools

DeepSeek ↗

DeepSeek V4.1 Flash39.1%

Text only

subset: text · No tools

DeepSeek ↗

Grok 4 Heavy50.7%

Text only

Heavy parallel test-time compute and native tool use. · subset: text · Tools enabled

xAI ↗

Muse Spark58%

Unspecified subset

Contemplating mode; parallel agents. Tool access not explicit in cited prose. · subset: unspecified

Meta ↗

Kimi K2.654%

Unspecified subset

Provider announcement; tools enabled; subset not explicitly named in announcement · Tools enabled

Moonshot AI ↗

DeepSeek V4 Pro 042448.2%

Unspecified subset

Original V4 Think Max; HLE with tools; source does not explicitly identify subset · Tools enabled

DeepSeek ↗

DeepSeek V4 Flash 042045.1%

Unspecified subset

Original V4 Think Max; HLE with tools; source does not explicitly identify subset · Tools enabled

DeepSeek ↗

DeepSeek V4.1 Flash63.9%

Unspecified subset

subset: unspecified · Tools enabled

DeepSeek ↗

GPT-5.6 Sol64.5%

Unspecified subset

temperature=1.0; top_p=0.95; max generation 163840 tokens; max context 300000 tokens; context management; GPT-5.6 Luna medium judge. · subset: unspecified · Tools enabled

Z.ai ↗

GLM-5.362.5%

Unspecified subset

temperature=1.0; top_p=0.95; max generation 163840 tokens; max context 300000 tokens; context management; GPT-5.6 Luna medium judge. · subset: unspecified · Tools enabled

Z.ai ↗

Kimi K359.8%

Unspecified subset

temperature=1.0; top_p=0.95; max generation 163840 tokens; max context 300000 tokens; context management; GPT-5.6 Luna medium judge. · subset: unspecified · Tools enabled

Z.ai ↗

DeepSeek V4 Pro 081360%

Unspecified subset

subset: unspecified · Tools enabled

DeepSeek ↗

DeepSeek V4 Pro 081342.7%

Unspecified subset

subset: unspecified · No tools

DeepSeek ↗

Other benchmarks

GDPval-AA

GDPval-AA evaluates models on professional knowledge-work tasks from OpenAI’s GDPval dataset, including creating documents, spreadsheets, slides, and other work products.

LiveBench: Research

LiveBench 2.0 evaluates research reports on its October 25, 2026 release.

π-Bench

π-Bench evaluates assistants across 100 multi-turn tasks in persistent workspaces.