All models ranked by average score across reasoning/math benchmarks (AIME 2025, ARC Challenge, GPQA Diamond, Humanity's Last Exam, LiveBench, MATH-500).

#	Model	Reasoning Avg	AIME 2025	ARC Challenge	GPQA Diamond	Quality	Price
1	GPT-5.2 Pro Reasoning OpenAI	94.8	—	—	91.0	93.0	$21.00
2	GPT-5 Pro OpenAI	92.8	—	—	88.0	90.0	$15.00
3	O4 Mini Reasoning OpenAI	90.1	92.7	—	81.4	90.0	$1.10
4	O3 Reasoning OpenAI	89.2	91.6	—	83.3	88.0	$2.00
5	o1 Reasoning OpenAI	87.2	—	—	78.0	84.0	$15.00
6	R1 0528 Reasoning OSS DeepSeek	87.1	87.5	—	76.0	83.0	$0.50
7	GPT-5 OpenAI	86.0	—	—	86.0	87.0	$1.25
8	Qwen3 235B A22B Reasoning OSS Alibaba	85.0	—	—	—	87.0	$0.46
9	o3 Mini Reasoning OpenAI	84.5	—	—	75.0	84.0	$1.10
10	Claude Sonnet 4.6 Reasoning Anthropic	83.0	—	—	—	86.0	$3.00
11	Qwen3 Max OSS Alibaba	82.8	—	—	72.0	85.0	$0.78
12	Gemini 2.5 Pro Reasoning Google	81.8	86.7	—	68.4	83.0	$1.25
13	Claude Opus 4 Anthropic	80.4	—	—	72.1	84.0	$15.00
14	Claude Opus 4.5 Anthropic	78.0	—	—	78.0	86.0	$5.00
15	O3 Pro Reasoning OpenAI	77.2	96.7	—	87.5	88.0	$20.00
16	Llama 4 Scout OSS Meta	76.8	—	—	—	79.0	$0.08
17	Claude Sonnet 4 Anthropic	76.5	—	—	67.5	79.0	$3.00
18	GPT-4.1 OpenAI	74.7	—	—	66.3	77.0	$2.00
19	DeepSeek V3.2 Reasoning OSS DeepSeek	73.0	—	—	62.0	77.0	$0.20
20	Claude Sonnet 4.5 Anthropic	72.0	—	—	72.0	79.0	$3.00
21	Gemini 2.5 Flash Reasoning Google	70.8	—	—	59.2	78.0	$0.30
22	GPT-4o-mini OpenAI	70.2	—	—	—	74.0	$0.15
23	DeepSeek R1 Reasoning OSS DeepSeek	69.3	79.8	—	71.5	85.0	$0.70
24	Claude 3.5 Haiku Anthropic	69.3	—	—	—	76.0	$0.80
25	Llama 4 Maverick OSS Meta	68.8	—	—	56.0	76.0	$0.15
26	DeepSeek V3 Reasoning OSS DeepSeek	68.7	—	—	59.1	76.0	$0.23
27	Phi 4 Reasoning OSS Microsoft	68.3	—	—	56.1	74.0	$0.07
28	GPT-5.2 OpenAI	67.0	—	—	89.0	90.0	$1.75
29	GPT-4o (2024-05-13) OpenAI	65.1	—	—	53.6	75.0	$5.00
30	Qwen2.5 72B Instruct OSS Alibaba	64.5	—	—	49.0	71.0	$0.36
31	Mistral Large OSS Mistral	58.0	—	—	58.0	73.0	$2.00
32	Claude Opus 4.6 Reasoning Anthropic	54.3	—	—	—	89.0	$15.00
33	Llama 3.3 70B Instruct OSS Meta	50.7	—	—	50.7	71.0	$0.10

What are reasoning models?

Reasoning models (like OpenAI's o-series and DeepSeek R1) use extended "chain-of-thought" processing to work through complex problems step by step. They're particularly strong at:

Mathematics: Competition-level math problems (AIME, MATH-500)
Science: Graduate-level science questions (GPQA Diamond)
Coding: Complex software engineering tasks (SWE-Bench)
Logic: Multi-step logical deduction and constraint satisfaction

The trade-off is higher latency and cost — reasoning models "think" before responding, which takes longer but produces more accurate answers for hard problems.

Coding models → Full leaderboard → Compare head-to-head →