Best LLMs for advanced knowledge and reasoning

HLE-Diamond

HLE-Diamond is a revised 1,000-question version of Humanity’s Last Exam, split into 500 reasoning and 500 knowledge questions. These results use the official prompts without tools. Scores are separate from the original Humanity’s Last Exam.

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

Widely usedNot saturated

Official benchmark

HLE-Diamond 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
Gemini 4 CarbonGoogle · Proprietary	1,000 questions / no tools	69.1%
Claude Opus 5.5Anthropic · Proprietary	1,000 questions / no tools	67.9%
GPT-6 AstraOpenAI · Proprietary	1,000 questions / no tools	63.8%
Claude Fable 5.1Anthropic · Proprietary	1,000 questions / no tools	60.1%
GPT-6.1 SolOpenAI · Proprietary	1,000 questions / no tools	58.8%
Gemini 4 ArgonGoogle · Proprietary	1,000 questions / no tools	56.7%
Claude Sonnet 5.5Anthropic · Proprietary	1,000 questions / no tools	53.3%
GPT-6.1 TerraOpenAI · Proprietary	1,000 questions / no tools	47.3%
Muse Wonder 1.0Meta · Proprietary	1,000 questions / no tools	39%
Kimi K3.1 ProMoonshot AI · Open source / open weights	1,000 questions / no tools	37.3%
Muse Spark 1.3Meta · Proprietary	1,000 questions / no tools	33.7%
Gemini 3.8 FlashGoogle · Proprietary	1,000 questions / no tools	33.1%
Gemini 3.1 ProGoogle · Proprietary	1,000 questions / no tools	31.4%
Claude Haiku 5.5Anthropic · Proprietary	1,000 questions / no tools	29.6%
GPT-5.6 TerraOpenAI · Proprietary	1,000 questions / no tools	28.8%
MiniMax M3.1 FlashMiniMax · Open source / open weights	1,000 questions / no tools	28.1%
Kimi K3Moonshot AI · Open source / open weights	1,000 questions / no tools	26.9%
Muse Spark OpenMeta · Open source / open weights	1,000 questions / no tools	26.3%
DeepSeek V4.1 Flash 1017DeepSeek · Open source / open weights	1,000 questions / no tools	25.3%
Step 5 PreviewStepFun · Open source / open weights	1,000 questions / no tools	24.3%
Grok 4.7xAI · Proprietary	1,000 questions / no tools	23.6%
GPT-6 LunaOpenAI · Proprietary	1,000 questions / no tools	22.2%
DeepSeek V4 Pro 0813DeepSeek · Open source / open weights	1,000 questions / no tools	19.5%
DeepSeek V4.1 FlashDeepSeek · Open source / open weights	1,000 questions / no tools	19.1%
GLM-5.3-FlashZ.ai · Open source / open weights	1,000 questions / no tools	18.2%
GLM-5.3Z.ai · Open source / open weights	1,000 questions / no tools	17.9%
MiniMax M3MiniMax · Open source / open weights	1,000 questions / no tools	16%
Qwen3.8 27BQwen · Open source / open weights	1,000 questions / no tools	15.3%
MiMo-V2.6-ProXiaomi · Open source / open weights	1,000 questions / no tools	15.2%
Qwen3.5-397B-A17BQwen · Open source / open weights	1,000 questions / no tools	12.4%
InklingThinking Machines · Open source / open weights	1,000 questions / no tools	10.6%
Nemotron 3.5 Super VLNVIDIA · Open source / open weights	1,000 questions / no tools	8.8%
gpt-oss-20bOpenAI · Open source / open weights	1,000 questions / no tools	4.8%
Qwen 3.5 9BQwen · Open source / open weights	1,000 questions / no tools	4.8%
gpt-oss-120bOpenAI · Open source / open weights	1,000 questions / no tools	3.8%

View all 35 published reports

Gemini 4 Carbon69.1%

1,000 questions / no tools

selection: best reported configuration per release · Gemini 4 Carbon (High) · No tools

Claude Opus 5.567.9%

1,000 questions / no tools

selection: best reported configuration per release · Claude Opus 5.5 (Max, Default Fallback) · No tools

GPT-6 Astra63.8%

1,000 questions / no tools

selection: best reported configuration per release · GPT-6 Astra (Max) · No tools

Claude Fable 5.160.1%

1,000 questions / no tools

selection: best reported configuration per release · Claude Fable 5.1 (Max, Default Fallback) · No tools

GPT-6.1 Sol58.8%

1,000 questions / no tools

selection: best reported configuration per release · GPT-6.1 Sol (Max) · No tools

Gemini 4 Argon56.7%

1,000 questions / no tools

selection: best reported configuration per release · Gemini 4 Argon (High) · No tools

Claude Sonnet 5.553.3%

1,000 questions / no tools

selection: best reported configuration per release · Claude Sonnet 5.5 (Max, Default Fallback) · No tools

GPT-6.1 Terra47.3%

1,000 questions / no tools

selection: best reported configuration per release · GPT-6.1 Terra (max) · No tools

Muse Wonder 1.039%

1,000 questions / no tools

selection: best reported configuration per release · Muse Wonder (max) · No tools

Kimi K3.1 Pro37.3%

1,000 questions / no tools

selection: best reported configuration per release · Kimi K3.1 Pro · No tools

Muse Spark 1.333.7%

1,000 questions / no tools

selection: best reported configuration per release · Muse Spark 1.3 (Max) · No tools

Gemini 3.8 Flash33.1%

1,000 questions / no tools

selection: best reported configuration per release · Gemini 3.8 Flash (High) · No tools

Gemini 3.1 Pro31.4%

1,000 questions / no tools

selection: best reported configuration per release · Gemini 3.1 Pro Preview · No tools

Claude Haiku 5.529.6%

1,000 questions / no tools

selection: best reported configuration per release · Claude Haiku 5.5 (max) · No tools

GPT-5.6 Terra28.8%

1,000 questions / no tools

selection: best reported configuration per release · GPT-5.6 Terra (max) · No tools

MiniMax M3.1 Flash28.1%

1,000 questions / no tools

selection: best reported configuration per release · MiniMax-M3.1-Flash · No tools

Kimi K326.9%

1,000 questions / no tools

selection: best reported configuration per release · Kimi K3 (Max) · No tools

Muse Spark Open26.3%

1,000 questions / no tools

selection: best reported configuration per release · Muse Spark Open (xhigh) · No tools

DeepSeek V4.1 Flash 101725.3%

1,000 questions / no tools

selection: best reported configuration per release · DeepSeek V4.1 Flash 1017 (Max) · No tools

Step 5 Preview24.3%

1,000 questions / no tools

selection: best reported configuration per release · Step 5 Preview · No tools

Grok 4.723.6%

1,000 questions / no tools

selection: best reported configuration per release · Grok 4.7 (High) · No tools

GPT-6 Luna22.2%

1,000 questions / no tools

selection: best reported configuration per release · GPT-6 Luna (max) · No tools

DeepSeek V4 Pro 081319.5%

1,000 questions / no tools

selection: best reported configuration per release · DeepSeek V4 Pro 0813 (Max) · No tools

DeepSeek V4.1 Flash19.1%

1,000 questions / no tools

selection: best reported configuration per release · DeepSeek V4.1 Flash (Max) · No tools

GLM-5.3-Flash18.2%

1,000 questions / no tools

selection: best reported configuration per release · GLM-5.3-Flash · No tools

GLM-5.317.9%

1,000 questions / no tools

selection: best reported configuration per release · GLM-5.3 (Max) · No tools

MiniMax M316%

1,000 questions / no tools

selection: best reported configuration per release · MiniMax-M3 · No tools

Qwen3.8 27B15.3%

1,000 questions / no tools

selection: best reported configuration per release · Qwen3.8 27B (xhigh) · No tools

MiMo-V2.6-Pro15.2%

1,000 questions / no tools

selection: best reported configuration per release · MiMo-V2.6-Pro · No tools

Qwen3.5-397B-A17B12.4%

1,000 questions / no tools

selection: best reported configuration per release · Qwen3.5 397B A17B · No tools

Inkling10.6%

1,000 questions / no tools

selection: best reported configuration per release · Inkling (xhigh) · No tools

Nemotron 3.5 Super VL8.8%

1,000 questions / no tools

selection: best reported configuration per release · Nemotron 3.5 Super VL · No tools

Qwen 3.5 9B4.8%

1,000 questions / no tools

selection: best reported configuration per release · Qwen3.5 9B · No tools

gpt-oss-20b4.8%

1,000 questions / no tools

selection: best reported configuration per release · gpt-oss-20b (High) · No tools

gpt-oss-120b3.8%

1,000 questions / no tools

selection: best reported configuration per release · gpt-oss-120b (High) · No tools

Other benchmarks

GDP.pdf

GDP.pdf tests reasoning over long professional documents across ten domains.

OSWorld

OSWorld evaluates computer-use agents on practical tasks in a desktop environment.

LiveBench: Design

LiveBench 2.0 evaluates document and presentation creation on its October 25, 2026 release.