AI model leaderboard

Every tracked model ranked across the headline benchmarks. The Intelligence Index averages each model's normalized scores; click any benchmark column header to sort by it.

Best overall

GPT 5.2 Thinking

90.8 Intelligence Index

Best at knowledge

GPT 5.4 Pro

94.4 GPQA Diamond

Best at math

GPT 5.2 Thinking

100.0 AIME 2025

Best at coding

Claude Opus 4.7

87.6 SWE-bench Verified

Best at multimodal

GPT 5.1

84.2 MMMU

Models × benchmarks

Cells are best primary scores. Color intensity reflects normalized score. Click a column header to sort.

#	Model	MMLU-Pro	GPQA Diamond	Humanity's Last Exam	AIME 2025	SWE-bench Verified	LiveCodeBench	MMMU	AA-LCR	Intelligence Index
1	GPT 5.2 Thinking	—	92.4%	—	100.0%	80.0%	—	—	—	90.8 3/8
2	Grok 3 Think	—	84.6%	—	93.3%	—	79.4%	—	—	85.8 3/8
3	GPT 5.1	—	88.1%	—	94.6%	74.9%	—	84.2%	—	85.5 4/8
4	Gemma 4	85.2%	84.3%	—	—	—	80.0%	—	—	83.2 3/8
5	Claude Opus 4.5	—	87.0%	—	—	80.9%	—	80.7%	—	82.9 3/8
6	Claude Sonnet 4.5	—	83.4%	—	87.0%	77.2%	—	77.8%	—	81.4 4/8
7	Qwen 3.5 27B	86.1%	85.5%	—	—	72.4%	—	—	—	81.3 3/8
8	o4 mini	—	—	—	92.7%	68.1%	—	81.6%	—	80.8 3/8
9	DeepSeek V3.2 Exp	85.0%	79.9%	—	89.3%	67.8%	74.1%	—	—	79.2 5/8
10	Kimi K2.6	—	90.5%	54.0%	—	80.2%	89.6%	—	—	78.6 4/8
11	Qwen 3.5 122B A10B	—	86.6%	—	—	72.0%	—	76.9%	—	78.5 3/8
12	Grok 4 Heavy	—	88.4%	44.4%	100.0%	—	79.4%	—	—	78.1 4/8
13	Nemotron 3 Nano	78.3%	75.0%	—	89.1%	—	68.3%	—	—	77.7 4/8
14	Deepseek 3.2	85.0%	82.4%	40.8%	93.1%	73.1%	83.3%	—	—	76.3 6/8
15	Claude Opus 4.7	—	94.2%	46.9%	—	87.6%	—	—	—	76.2 3/8
16	Gemini 3 Pro	—	91.9%	37.5%	95.0%	76.2%	—	—	—	75.2 4/8
17	Claude Haiku 4.5	—	73.0%	—	80.7%	73.3%	—	73.2%	—	75.1 4/8
18	Phi 4 reasoning plus	76.0%	69.3%	—	78.0%	—	—	—	—	74.4 3/8
19	Gemini 3 Flash (Thinking)	—	90.4%	33.7%	95.2%	78.0%	—	—	—	74.3 4/8
20	Claude Sonnet 4	—	75.4%	—	70.5%	72.7%	—	74.4%	—	73.3 4/8
21	Gemini 3.1 Pro	—	94.3%	44.4%	—	80.6%	—	—	—	73.1 3/8
22	Seed 1.5	80.1%	65.0%	—	—	—	—	73.9%	—	73.0 3/8
23	GPT 5 (Thinking)	—	85.7%	24.8%	94.6%	74.9%	—	84.2%	—	72.8 5/8
24	Claude Sonnet 3.7 (Thinking)	—	78.2%	—	—	62.3%	—	75.0%	—	71.8 3/8
25	Grok 3	79.9%	75.4%	—	—	—	57.0%	73.2%	—	71.4 4/8
26	Grok 4	—	87.5%	25.4%	91.7%	—	79.0%	—	—	70.9 4/8
27	Llama 4 Behemoth	82.2%	73.7%	—	—	—	49.4%	76.1%	—	70.4 4/8
28	DeepSeek V3.1 Terminus	85.0%	80.7%	21.7%	88.4%	—	74.9%	—	—	70.1 5/8
29	Muse Spark	—	89.5%	42.8%	—	77.4%	—	—	—	69.9 3/8
30	o3	—	83.3%	20.3%	88.9%	69.1%	—	82.9%	—	68.9 5/8
31	DeepSeek-R1	84.0%	71.5%	—	70.0%	49.2%	—	—	—	68.7 4/8
32	GLM 4.6	—	81.0%	17.2%	93.9%	68.0%	82.8%	—	—	68.6 5/8
33	Claude Sonnet 4.6	—	89.9%	33.2%	—	79.6%	—	—	—	67.6 3/8
34	Llama 4 Maverick	80.5%	69.8%	—	—	—	43.4%	73.4%	—	66.8 4/8
35	Qwen3 30B A3B	—	65.8%	—	70.9%	—	62.6%	—	—	66.4 3/8
36	Claude Sonnet 3.7	—	62.3%	—	—	62.3%	—	71.8%	—	65.5 3/8
37	GPT 4.1	—	66.3%	—	—	55.0%	—	75.0%	—	65.4 3/8
38	Gemini 2.5 Pro	—	84.0%	18.8%	86.7%	63.8%	70.4%	—	—	64.7 5/8
39	Grok 3 mini	78.9%	66.2%	—	—	—	41.5%	69.4%	—	64.0 4/8
40	Gemini 2.5 Pro (Thinking)	—	86.4%	21.6%	88.0%	59.6%	—	—	—	63.9 4/8
41	Magistral Medium	—	70.8%	—	64.9%	—	50.3%	—	—	62.0 3/8
42	Kimi K2 Instruct	—	75.1%	—	49.5%	65.8%	53.7%	—	—	61.0 4/8
43	DeepSeek V3	75.9%	59.1%	—	—	42.0%	—	—	—	59.0 3/8
44	Llama 4 Scout	74.3%	57.2%	—	—	—	32.8%	69.4%	—	58.4 4/8
45	o1	—	78.0%	8.12%	79.2%	48.9%	—	77.6%	—	58.4 5/8
46	Gemini 2.5 Flash (Thinking)	—	82.8%	11.0%	72.0%	60.4%	—	—	—	56.6 4/8
47	GPT 5	—	77.8%	6.30%	61.9%	52.8%	—	74.4%	—	54.6 5/8
48	Claude Haiku 3.5	41.6%	65.0%	—	—	40.6%	—	—	—	49.1 3/8
49	Gemini 2.5 Flash-Lite	—	64.6%	5.10%	49.8%	31.6%	33.7%	72.9%	—	43.0 6/8
50	GPT 5.5 Instant	—	—	—	81.2%	—	—	—	—	—
51	Mistral Medium 3.5	—	—	—	—	77.6%	—	—	—	—
52	Deepseek V4 Pro	—	—	—	—	—	93.5%	—	—	—
53	GPT 5.5	—	93.6%	41.4%	—	—	—	—	—	—
54	GLM-5.1	—	86.2%	31.0%	—	—	—	—	—	—
55	GPT 5.4 Pro	—	94.4%	—	—	—	—	—	—	—
56	GPT 5.4	—	92.8%	—	—	57.7%	—	—	—	—
57	Gemma 3	78.0%	72.6%	—	—	—	—	—	—	—
58	GPT-4 Turbo	—	50.4%	—	—	—	—	—	—	—
59	GPT 5.1 Thinking	—	88.1%	—	94.6%	—	—	—	—	—
60	Claude Opus 3	—	50.4%	—	—	—	—	—	—	—
61	Mistral Small 3	66.3%	—	—	—	—	—	—	—	—
62	Pixtral Large	—	—	—	—	—	—	64.0%	—	—
63	GPT 5.4 Mini	—	88.0%	—	—	—	—	—	—	—
64	GPT 5.4 Nano	—	82.8%	—	—	—	—	—	—	—
65	Qwen 3.5 35B A3B	—	84.2%	—	—	69.2%	—	—	—	—
66	Nova Premier	—	—	—	—	42.4%	—	—	—	—
67	Command A	69.6%	50.8%	—	—	—	—	—	—	—
68	Qwen3 Coder	—	—	—	—	67.0%	—	—	—	—
69	Claude Opus 4.6	—	91.3%	—	—	80.8%	—	—	—	—
70	GPT 5.2 Pro	—	93.2%	—	—	—	—	—	—	—
71	Gemini 3 Flash	—	90.4%	—	—	78.0%	—	—	—	—
72	Gemini 3 Deep Think	—	93.8%	41.0%	—	—	—	—	—	—
73	MiniMax M2.5	—	—	—	—	80.2%	—	—	—	—
74	GLM 5	—	86.0%	—	—	77.8%	—	—	—	—
75	Kimi K2.5	—	—	—	—	76.8%	85.0%	—	—	—
76	DeepSeek 3.2 Speciale	—	—	30.6%	96.0%	—	—	—	—	—
77	Devstral 2	—	—	—	—	72.2%	—	—	—	—
78	Mistral Large 3	—	43.9%	—	—	—	34.4%	—	—	—
79	GPT OSS 120B	90.0%	80.1%	—	—	—	—	—	—	—
80	Nemotron 3 Super	75.7%	60.0%	—	—	—	—	—	—	—
81	Qwen3-235B-A22B	—	—	—	81.5%	—	70.7%	—	—	—
82	Qwen3-30B-A3B	—	65.8%	—	—	—	—	—	—	—
83	R1 1776	—	71.5%	—	70.0%	—	—	—	—	—
84	Grok Code Fast 1	—	—	—	—	70.8%	—	—	—	—
85	Llama 3.1 Nemotron Ultra	—	76.0%	—	—	—	—	—	—	—
86	Opus 4.1 Thinking	—	80.9%	—	—	74.5%	—	—	—	—
87	Nova Pro	—	46.9%	—	—	—	—	—	—	—
88	Nova Lite	—	42.0%	—	—	—	—	—	—	—
89	Nova Micro	—	40.0%	—	—	—	—	—	—	—
90	GPT 5.3 Codex	—	92.6%	—	—	56.8%	—	—	—	—
91	Llama 3.2	—	32.8%	—	—	—	—	—	—	—
92	Pixtral 12B	—	—	—	—	—	—	52.0%	—	—
93	GPT-4o	—	53.6%	—	—	—	—	69.1%	—	—
94	Qwen3 235B A22B	—	—	—	81.5%	—	70.7%	—	—	—
95	Llama 3.3	68.9%	50.5%	—	—	—	—	—	—	—

Capability scatter

Each dot is a model. Position shows two-axis capability; size reflects how many headline benchmarks the model has been scored on.

X axis Y axis

Go to section

Search

AI model leaderboard

Models × benchmarks

Capability scatter

Help

People also viewed

Create AI Tools

Mini Tool

Vibe code an AI Tool

Choose listing type: