LLM Leaderboard

Ranked by AIME · source AA

American Invitational Mathematics Examination — olympiad math.

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

#1 by AIME

GPT-5 (high)

95.7%

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

#	Model	Creator	AIME	In $/1M	Out $/1M	tok/s	LMArena
1	GPT-5 (high)	OpenAI	95.7%	$1.25	$10.0	110	1405
2	Grok 4	SpaceXAI	94.3%	$5.50	$27.5	0	1452
3	o4-mini (high)	OpenAI	94.0%	$1.10	$4.40	180	—
4	Qwen3 235B A22B 2507 (Reasoning)	Alibaba	94.0%	$0.70	$8.40	62	1366
5	Grok 3 mini Reasoning (high)	SpaceXAI	93.3%	$0.30	$0.50	49	—
6	GPT-5 (medium)	OpenAI	91.7%	$1.25	$10.0	82	—
7	Qwen3 30B A3B 2507 (Reasoning)	Alibaba	90.7%	$0.20	$2.40	154	1317
8	o3	OpenAI	90.3%	$2.00	$8.00	135	—
9	DeepSeek R1 0528 (May '25)	DeepSeek	89.3%	$1.35	$4.20	0	1373
10	Gemini 2.5 Pro	Google	88.7%	$1.25	$10.0	145	1457
11	GLM-4.5 (Reasoning)	Z AI	87.3%	$0.00	$0.00	0	1429
12	Gemini 2.5 Pro Preview (Mar' 25)	Google	87.0%	$0.00	$0.00	0	1457
13	o3-mini (high)	OpenAI	86.0%	$1.10	$4.40	247	1337
14	Llama Nemotron Super 49B v1.5 (Reasoning)	NVIDIA	86.0%	$0.40	$0.40	53	—
15	MiniMax M1 80k	MiniMax	84.7%	$0.55	$2.20	0	1343
16	Gemini 2.5 Pro Preview (May' 25)	Google	84.3%	$1.25	$10.0	0	1457
17	Gemini 2.5 Flash Preview (Reasoning)	Google	84.3%	$0.00	$0.00	0	1417
18	EXAONE 4.0 32B (Reasoning)	LG AI Research	84.3%	$0.00	$0.00	0	—
19	Qwen3 235B A22B (Reasoning)	Alibaba	84.0%	$0.70	$8.40	63	1366
20	GPT-5 (low)	OpenAI	83.0%	$1.25	$10.0	83	—
21	Gemini 2.5 Flash (Reasoning)	Google	82.3%	$0.30	$2.50	232	1417
22	MiniMax M1 40k	MiniMax	81.3%	$0.00	$0.00	0	1343
23	Qwen3 32B (Reasoning)	Alibaba	80.7%	$0.70	$8.40	98	1340
24	Sonar Reasoning Pro	Perplexity	79.0%	$0.00	$0.00	0	—
25	QwQ 32B	Alibaba	78.0%	$0.66	$1.00	32	1329
26	Claude 4 Sonnet (Reasoning)	Anthropic	77.3%	$3.00	$15.0	0	—
27	o3-mini	OpenAI	77.0%	$1.10	$4.40	224	1319
28	Sonar Reasoning	Perplexity	77.0%	$0.00	$0.00	0	—
29	Qwen3 14B (Reasoning)	Alibaba	76.3%	$0.35	$4.20	63	—
30	Claude 4 Opus (Reasoning)	Anthropic	75.7%	$15.0	$75.0	0	—
31	Qwen3 30B A3B (Reasoning)	Alibaba	75.3%	$0.20	$2.40	113	1317
32	Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)	NVIDIA	74.7%	$0.60	$1.80	54	1320
33	Qwen3 8B (Reasoning)	Alibaba	74.7%	$0.18	$2.10	63	—
34	Qwen3 30B A3B 2507 Instruct	Alibaba	72.7%	$0.20	$0.80	160	1317
35	o1	OpenAI	72.3%	$15.0	$60.0	126	—
36	Qwen3 235B A22B 2507 Instruct	Alibaba	71.7%	$0.70	$2.80	66	1419
37	Magistral Small 1	Mistral	71.3%	$0.00	$0.00	0	—
38	Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)	NVIDIA	70.7%	$0.00	$0.00	0	—
39	Gemini 2.5 Flash-Lite (Reasoning)	Google	70.3%	$0.10	$0.40	307	1417
40	Magistral Medium 1	Mistral	70.0%	$0.00	$0.00	0	—
41	Kimi K2	Kimi	69.3%	$0.57	$2.30	36	1455
42	Solar Pro 2 (Reasoning)	Upstage	69.0%	$0.00	$0.00	0	—
43	DeepSeek R1 Distill Qwen 32B	DeepSeek	68.7%	$0.00	$0.00	0	1373
44	DeepSeek R1 (Jan '25)	DeepSeek	68.3%	$1.68	$4.70	0	1373
45	GLM-4.5-Air	Z AI	67.3%	$0.17	$0.98	73	1383
46	DeepSeek R1 Distill Llama 70B	DeepSeek	67.0%	$0.70	$1.05	24	1373
47	DeepSeek R1 Distill Qwen 14B	DeepSeek	66.7%	$0.00	$0.00	0	1373
48	Solar Pro 2 (Preview) (Reasoning)	Upstage	66.3%	$0.00	$0.00	0	—
49	Qwen3 4B (Reasoning)	Alibaba	65.7%	$0.00	$0.00	0	—
50	DeepSeek R1 0528 Qwen3 8B	DeepSeek	65.0%	$0.00	$0.00	0	1428

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.