LLM Leaderboard

Ranked by MATH-500 · source AA

Subset of the MATH benchmark.

IntelligenceAA LMArenaLMArena CodingAA MathAA MMLU-ProAA GPQAAA LiveCodeAA MATH-500AA AIMEAA HLEAA SciCodeAA

#1 by MATH-500

GPT-5 (high)

99.4%

OpenAI · in $1.25 / out $10.0 per 1M · 110 tok/s

#	Model	Creator	MATH-500	In $/1M	Out $/1M	tok/s	LMArena
1	GPT-5 (high)	OpenAI	99.4%	$1.25	$10.0	110	1405
2	o3	OpenAI	99.2%	$2.00	$8.00	135	—
3	Grok 3 mini Reasoning (high)	SpaceXAI	99.2%	$0.30	$0.50	49	—
4	GPT-5 (medium)	OpenAI	99.1%	$1.25	$10.0	82	—
5	Claude 4 Sonnet (Reasoning)	Anthropic	99.1%	$3.00	$15.0	0	—
6	Grok 4	SpaceXAI	99.0%	$5.50	$27.5	0	1452
7	o4-mini (high)	OpenAI	98.9%	$1.10	$4.40	180	—
8	GPT-5 (low)	OpenAI	98.7%	$1.25	$10.0	83	—
9	Gemini 2.5 Pro Preview (May' 25)	Google	98.6%	$1.25	$10.0	0	1457
10	o3-mini (high)	OpenAI	98.5%	$1.10	$4.40	247	1337
11	Qwen3 235B A22B 2507 (Reasoning)	Alibaba	98.4%	$0.70	$8.40	62	1366
12	Llama Nemotron Super 49B v1.5 (Reasoning)	NVIDIA	98.3%	$0.40	$0.40	53	—
13	DeepSeek R1 0528 (May '25)	DeepSeek	98.3%	$1.35	$4.20	0	1373
14	Claude 4 Opus (Reasoning)	Anthropic	98.2%	$15.0	$75.0	0	—
15	Gemini 2.5 Flash (Reasoning)	Google	98.1%	$0.30	$2.50	232	1417
16	Gemini 2.5 Flash Preview (Reasoning)	Google	98.1%	$0.00	$0.00	0	1417
17	Gemini 2.5 Pro Preview (Mar' 25)	Google	98.0%	$0.00	$0.00	0	1457
18	Qwen3 235B A22B 2507 Instruct	Alibaba	98.0%	$0.70	$2.80	66	1419
19	MiniMax M1 80k	MiniMax	98.0%	$0.55	$2.20	0	1343
20	GLM-4.5 (Reasoning)	Z AI	97.9%	$0.00	$0.00	0	1429
21	EXAONE 4.0 32B (Reasoning)	LG AI Research	97.7%	$0.00	$0.00	0	—
22	Qwen3 30B A3B 2507 (Reasoning)	Alibaba	97.6%	$0.20	$2.40	154	1317
23	Qwen3 30B A3B 2507 Instruct	Alibaba	97.5%	$0.20	$0.80	160	1317
24	o3-mini	OpenAI	97.3%	$1.10	$4.40	224	1319
25	MiniMax M1 40k	MiniMax	97.2%	$0.00	$0.00	0	1343
26	Kimi K2	Kimi	97.1%	$0.57	$2.30	36	1455
27	o1	OpenAI	97.0%	$15.0	$60.0	126	—
28	Gemini 2.5 Flash-Lite (Reasoning)	Google	96.9%	$0.10	$0.40	307	1417
29	Gemini 2.5 Pro	Google	96.7%	$1.25	$10.0	145	1457
30	Solar Pro 2 (Reasoning)	Upstage	96.7%	$0.00	$0.00	0	—
31	DeepSeek R1 (Jan '25)	DeepSeek	96.6%	$1.68	$4.70	0	1373
32	GLM-4.5-Air	Z AI	96.5%	$0.17	$0.98	73	1383
33	Magistral Small 1	Mistral	96.3%	$0.00	$0.00	0	—
34	Qwen3 14B (Reasoning)	Alibaba	96.1%	$0.35	$4.20	63	—
35	Qwen3 32B (Reasoning)	Alibaba	96.1%	$0.70	$8.40	98	1340
36	Qwen3 30B A3B (Reasoning)	Alibaba	95.9%	$0.20	$2.40	113	1317
37	Llama 3.3 Nemotron Super 49B v1 (Reasoning)	NVIDIA	95.9%	$0.00	$0.00	0	—
38	Sonar Reasoning Pro	Perplexity	95.7%	$0.00	$0.00	0	—
39	QwQ 32B	Alibaba	95.7%	$0.66	$1.00	32	1329
40	R1 1776	Perplexity	95.4%	$0.00	$0.00	0	—
41	Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)	NVIDIA	95.2%	$0.60	$1.80	54	1320
42	DeepSeek R1 Distill Qwen 14B	DeepSeek	94.9%	$0.00	$0.00	0	1373
43	Claude 3.7 Sonnet (Reasoning)	Anthropic	94.7%	$0.00	$0.00	0	—
44	Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)	NVIDIA	94.7%	$0.00	$0.00	0	—
45	o1-mini	OpenAI	94.4%	$0.00	$0.00	0	1317
46	Gemini 2.0 Flash Thinking Experimental (Jan '25)	Google	94.4%	$0.00	$0.00	0	—
47	Qwen3 Coder 480B A35B Instruct	Alibaba	94.2%	$1.50	$7.50	56	1356
48	DeepSeek V3 0324	DeepSeek	94.2%	$1.14	$1.25	0	1375
49	Claude 4 Opus (Non-reasoning)	Anthropic	94.1%	$15.0	$75.0	0	—
50	DeepSeek R1 Distill Qwen 32B	DeepSeek	94.1%	$0.00	$0.00	0	1373

Quality, pricing, and speed benchmarks via Artificial Analysis. Human preference Elo via LMArena. Refreshed daily at 06:00 UTC · last sync Sun, 19 Jul 2026 06:00:01 GMT.

About the metrics

Intelligence Index

Artificial Analysis composite of reasoning, knowledge, coding, math.

LMArena Elo

LMArena

Human preference Elo from chatbot arena head-to-head votes.

Coding Index

AA composite — LiveCodeBench, SciCode, etc.

Math Index

AA composite — MATH-500, AIME, etc.

MMLU-Pro

Multi-domain knowledge & reasoning. Harder than MMLU.

GPQA Diamond

Graduate-level science Q&A. Strong reasoning signal.

LiveCodeBench

Competitive programming, contamination-resistant.

MATH-500

Subset of the MATH benchmark.

AIME

American Invitational Mathematics Examination — olympiad math.

Humanity's Last Exam

Frontier reasoning benchmark; most models still score low.

SciCode

Scientific coding tasks.