Best LLM for Reasoning in 2026

Ranked by composite reasoning score averaging GPQA, MATH, AIME 2025, and AIME 2024. These benchmarks test graduate-level science, competition mathematics, and multi-step logical reasoning.

Updated automatically as new models release. Full benchmark leaderboard →

Gemini 3.1 Pro Google DeepMind

95.1%

avg

GPQA: 94.3%

MATH: 91.1%

AIME 2025: 100.0%

Grok 4 xAI

93.6%

avg

GPQA: 87.5%

MATH: 98.1%

AIME 2025: 95.1%

o3 OpenAI

93.2%

avg

GPQA: 87.7%

MATH: 96.7%

AIME 2025: 96.7%

AIME 2024: 91.6%

o4-mini OpenAI

91.4%

avg

GPQA: 81.4%

MATH: 99.5%

AIME 2025: 93.4%

Claude Opus 4.6 Anthropic

91.2%

avg

GPQA: 91.3%

AIME 2025: 91.0%

Gemini 2.5 Pro Google DeepMind

89.9%

avg

GPQA: 84.0%

MATH: 97.0%

AIME 2025: 86.7%

AIME 2024: 92.0%

Claude Sonnet 4.6 Anthropic

89.9%

avg

GPQA: 89.9%

Grok 3 xAI

89.8%

avg

GPQA: 84.6%

MATH: 97.6%

AIME 2025: 93.3%

AIME 2024: 83.9%

Claude Opus 4.5 Anthropic

88.0%

avg

GPQA: 87.0%

AIME 2025: 89.0%

DeepSeek R1 (0528)DeepSeek

86.6%

avg

GPQA: 81.0%

AIME 2025: 87.5%

AIME 2024: 91.4%

o1-pro OpenAI

86.0%

avg

GPQA: 78.3%

MATH: 96.4%

AIME 2024: 83.3%

Claude Sonnet 4.5 Anthropic

85.2%

avg

GPQA: 83.4%

AIME 2025: 87.0%

Gemini 2.5 Flash Google DeepMind

85.2%

avg

GPQA: 74.4%

MATH: 95.9%

Claude Opus 4 Anthropic

84.0%

avg

GPQA: 74.9%

MATH: 88.0%

AIME 2025: 89.0%

DeepSeek-R1 DeepSeek

79.7%

avg

GPQA: 71.5%

MATH: 97.3%

AIME 2025: 70.0%

AIME 2024: 79.8%

QwQ-32B Alibaba / Qwen

78.7%

avg

GPQA: 65.2%

MATH: 90.6%

AIME 2025: 79.5%

AIME 2024: 79.5%

Gemini 2.0 Flash Google DeepMind

78.3%

avg

GPQA: 67.0%

MATH: 89.7%

Qwen3 72B Alibaba / Qwen

76.9%

avg

GPQA: 65.5%

MATH: 92.2%

AIME 2025: 72.9%

Llama 4 Maverick Meta AI

75.8%

avg

GPQA: 69.8%

MATH: 81.7%

GPT-4.1 OpenAI

75.7%

avg

GPQA: 66.3%

MATH: 85.0%

Grok 3 mini xAI

67.4%

avg

GPQA: 67.4%

Phi-4 Microsoft

66.3%

avg

GPQA: 52.3%

MATH: 80.4%

GPT-4o OpenAI

65.1%

avg

GPQA: 53.6%

MATH: 76.6%

Claude 3.7 Sonnet Anthropic

61.2%

avg

GPQA: 77.2%

MATH: 87.5%

AIME 2025: 56.7%

AIME 2024: 23.3%

Mistral Small 3.1 Mistral AI

57.6%

avg

GPQA: 46.0%

MATH: 69.3%

GPT-4.1 mini OpenAI

50.3%

avg

GPQA: 50.3%

Claude Haiku 4.5 Anthropic

43.2%

avg

GPQA: 65.7%

MATH: 70.9%

AIME 2025: 20.0%

AIME 2024: 16.0%

What makes a good reasoning model?

Reasoning benchmarks test whether a model can solve multi-step problems requiring planning, logic, and domain knowledge — not just pattern matching or retrieval.

GPQA (Diamond) — Questions written by PhD-level experts in biology, chemistry, and physics. Designed so that non-experts who Google the answer still fail. The gold standard for deep scientific reasoning.
MATH — Competition mathematics at AMC/AIME difficulty. Tests multi-step algebraic and geometric reasoning.
AIME 2025 — The American Invitational Math Exam, 2025 edition. 30 hard problems, integer answers. Most recent math benchmark — 2025 numbers are resistant to training data contamination.
AIME 2024 — Same format, one year earlier. Used alongside 2025 to give a more stable picture of math reasoning capability.

For tasks involving complex analysis, research, legal and financial reasoning, or scientific work, a high GPQA score is the best predictor of real-world performance.

Also see: Best Coding LLM, Best Cheap LLM, Compare any two models.