基准测试类别

Math 基准测试

Compare top models across the benchmark suite that best represents math performance. Use this page as the fastest way to inspect the relevant tests, then jump into the full matrix when you want broader context.

类别中的基准测试

有覆盖的模型

有人类基准的基准测试

饱和的基准测试

包含内容

The current benchmark set in this category, with context on what each test captures.

所有基准测试

MATH Benchmark (500-problem subset)

Tests genuine mathematical reasoning, not just pattern matching. Reasoning models (o1, R1) dramatically outperform standard models here.

accuracy %

American Invitational Mathematics Examination 2024

Tests mathematical reasoning at competition level. Reasoning models achieve 70-90% while standard models struggle below 30%. Best differentiator for math ability.

accuracy %

Grade School Math 8K

饱和

Useful baseline for math ability, but now saturated — top models exceed 95%. More useful for evaluating smaller or open-source models.

accuracy %

Model type:

MATH Benchmark (500-problem subset)

Competition-level mathematics across algebra, geometry, number theory, counting/probability, intermediate algebra, and precalculus. 500-problem subset used by OpenAI for consistent evaluation.

Metric: accuracy %

Why it matters

Tests genuine mathematical reasoning, not just pattern matching. Reasoning models (o1, R1) dramatically outperform standard models here.

MATH-500 Scores (49 models)

Standard Reasoning Hybrid

LMMarketCap.com

#	Model	Provider	Type	Score	Tier
1	🥇o3	OpenAI	reasoning	99.0%	Elite
2	🥈o3-mini	OpenAI	reasoning	97.9%	Elite
3	🥉DeepSeek R1-0528	DeepSeek	reasoning	97.8%	Elite
4	DeepSeek R1	DeepSeek	reasoning	97.3%	Elite
5	o4-mini	OpenAI	reasoning	97.3%	Elite
6	o1	OpenAI	reasoning	96.4%	Elite
7	Gemini 3 Pro	Google	hybrid	96.0%	Elite
8	GPT-5.4	OpenAI	standard	95.5%	Elite
9	Gemini 2.5 Pro	Google	hybrid	95.2%	Elite
10	Grok 4	xAI	reasoning	95.0%	Elite
11	o1 Preview	OpenAI	reasoning	94.8%	Elite
12	GPT-5.2	OpenAI	standard	94.0%	Elite
13	GPT-5.1	OpenAI	standard	93.5%	Elite
14	GPT-5	OpenAI	standard	92.5%	Elite
15	DeepSeek V3 (March 2025)	DeepSeek	standard	92.0%	Elite
16	Claude Opus 4.6	Anthropic	hybrid	90.5%	Elite
17	DeepSeek V3	DeepSeek	standard	90.2%	Elite
18	o1-mini	OpenAI	reasoning	90.0%	Elite
19	Gemini 2.0 Flash	Google	standard	89.7%	Strong
20	Claude Opus 4.5	Anthropic	hybrid	88.1%	Strong
21	Gemini 3 Flash	Google	hybrid	88.0%	Strong
22	Claude Opus 4	Anthropic	hybrid	86.0%	Strong
23	Gemini 2.5 Flash	Google	hybrid	85.8%	Strong
24	Claude Sonnet 4.6	Anthropic	hybrid	85.3%	Strong
25	Grok 3	xAI	standard	85.0%	Strong
26	Qwen 2.5 Coder 32B	Alibaba	standard	83.5%	Strong
27	Qwen 2.5 72B	Alibaba	standard	83.1%	Strong
28	Claude Sonnet 4.5	Anthropic	hybrid	83.0%	Strong
29	Claude 3.7 Sonnet	Anthropic	hybrid	82.2%	Strong
30	Claude Sonnet 4	Anthropic	hybrid	81.4%	Strong
31	Llama 4 Maverick	Meta	standard	81.0%	Strong
32	Phi-4	Microsoft	standard	80.4%	Strong
33	GPT-4.1	OpenAI	standard	78.5%	Strong
34	Claude 3.5 Sonnet	Anthropic	standard	78.3%	Strong
35	Llama 3.3 70B	Meta	standard	77.0%	Strong
36	GPT-4o	OpenAI	standard	76.6%	Strong
37	Grok 2	xAI	standard	76.1%	Strong
38	Mistral Large 2	Mistral AI	standard	76.0%	Strong
39	Mistral Large 2	Mistral AI	standard	76.0%	Strong
40	Llama 3.1 405B	Meta	standard	73.8%	Good
41	GPT-4 Turbo	OpenAI	standard	72.6%	Good
42	Claude Haiku 4.5	Anthropic	standard	72.5%	Good
43	GPT-4o mini	OpenAI	standard	70.2%	Good
44	Claude 3.5 Haiku	Anthropic	standard	69.2%	Good
45	Llama 3.1 70B	Meta	standard	68.0%	Good
46	Gemini 1.5 Pro	Google	standard	67.7%	Good
47	Claude 3 Opus	Anthropic	standard	60.1%	Good
48	Mixtral 8x22B	Mistral AI	standard	60.0%	Good
49	Llama 4 Scout	Meta	standard	50.3%	Good

How to Read This Page

Performance Tiers

Elite - Top 10% of the score range

Strong - Top 25% of the score range

Good - Above the midpoint

Below Average - Below the midpoint

Model Types

Standard - Direct inference, no chain-of-thought

Reasoning - Extended thinking (o1, R1) - slower but excels on math/reasoning

Hybrid - Optional thinking mode (Claude 3.7, Gemini 2.5) - can switch between fast and deep

Saturated benchmarks have top models clustered above 90%, making them less useful for comparison.

Scores sourced from official model cards, technical reports, and third-party evaluations (Artificial Analysis, LMSYS Arena). Last updated: 2026-08-08T06:30:09.792Z. Some scores are approximate.

Frequently Asked Questions

AI benchmarks are grouped into categories like coding, math, reasoning, knowledge, and safety. Each category contains multiple standardized tests that measure specific aspects of model performance. This page focuses on one category so you can compare models within a specific skill area.

Each benchmark has its own scoring method - accuracy percentage, pass rate, Elo rating, or normalized score. We display raw scores from official evaluations and community-run tests. Scores are updated hourly as new evaluation results become available.

A saturated benchmark is one where top models score near the maximum (typically above 95%). This means the benchmark no longer effectively differentiates between the best models, and newer, harder benchmarks are needed to measure progress.