MATH and GSM8K.

Best AI models for math.

MATH (competition-level problems, formal proofs) weighted heaviest, GSM8K (grade-school word problems) as the floor. Models that win both handle algebra, calculus, and chain-of-thought arithmetic.

Benchmarks used: MATH · 70% GSM8K · 30%

#	Model	Score	From
1	DeepSeek R1 open	97.3	DeepSeek
2	GPT-5 closed	96.0	OpenAI
3	DeepSeek R1 Distill Llama 70B open	94.5	DeepSeek
4	DeepSeek R1 Distill Qwen 32B open	94.3	DeepSeek
5	DeepSeek R1 Distill Qwen 14B open	93.9	DeepSeek
6	Grok 3 closed	93.3	xAI
7	DeepSeek R1 Distill Qwen 7B open	92.8	DeepSeek
8	Gemini 2.5 Pro closed	92.0	Google DeepMind
9	DeepSeek V3 open	90.2	DeepSeek
10	Gemma 3 27B open	89.0	Google DeepMind
11	Claude Opus 4.7 closed	87.5	Anthropic
12	DeepSeek R1 Distill Qwen 1.5B open	83.9	DeepSeek
13	Qwen 2.5 72B open	83.1	Alibaba (Qwen Team)
14	Claude Sonnet 4.6 closed	82.0	Anthropic
15	Llama 3.3 70B open	77.0	Meta AI
16	GPT-4o closed	76.6	OpenAI
17	Llama 3.1 405B open	73.8	Meta AI
18	Mistral Large 2 open	73.0	Mistral AI
19	Mixtral 8x22B open	41.8	Mistral AI

Showing top 19 models with published data on at least one of the benchmarks above. Scores are weighted averages on a 0–100 scale.

AI model leaderboards

More leaderboards.

Best AI models for coding → Best AI models for reasoning → Best AI models for general knowledge → Best AI models for instruction-following → Best AI models for vision → Cheapest capable AI models →