modelbenchmark.io

Ranked by composite score

AI model leaderboard

One number per model, built from 16 published benchmarks across 4 sources. Each benchmark is weighted by how much of its spread survives its own measurement error, so a benchmark that no longer separates models stops counting. How the weights are computed. The quality rank uses scores only. Best per dollar is quality over output price. Speed is not in any rank. We do not sample live APIs.

Scope
Rank by
Weights
Price
Evidence
View

91 models ranked · 3 or more benchmarks, from 2 or more sources · percentile is out of all 344 scored models either way

RankModelLabCompositeEvidenceEffort rangeOut $/M
1GPT-6 AstraOpenAI99th15 bench · 2 src9.8 pts$50.00
2GPT-6.1 SolOpenAI99th13 bench · 2 src0.7 pts$10.00
3Claude Fable 5Anthropic98th15 bench · 2 src7.1 pts$50.00
4Claude Opus 5.5Anthropic95th15 bench · 2 src1.1 pts$20.00
5Claude Opus 5Anthropic93rd10 bench · 2 src11.0 pts$25.00
6Claude Fable 5.1Anthropic92nd14 bench · 2 src—$50.00
7GPT-6 SolOpenAI92nd13 bench · 2 src—$10.00
8GPT-5.6 SolOpenAI91st13 bench · 2 src16.2 pts$20.00
9Gemini 3 Pro PreviewGoogle90th5 bench · 2 src6.2 pts$12.00
10GPT-5.5OpenAI89th17 bench · 2 src21.8 pts$30.00
11Claude Sonnet 5.5Anthropic89th13 bench · 2 src2.4 pts$10.00
12Gemini 3.7 FlashGoogle89th9 bench · 2 src—$7.50
13Qwen3.8 MaxAlibaba88th8 bench · 2 src—$6.00
14Gemini 3.8 FlashGoogle88th9 bench · 2 src—$7.50
15DeepSeek V4 Pro 0813DeepSeek86th8 bench · 2 src—$3.96
16Claude Opus 4.8Anthropic86th14 bench · 2 src3.8 pts$25.00
17GPT-5.6 TerraOpenAI85th9 bench · 2 src15.4 pts$12.00
18GPT-5.4OpenAI85th14 bench · 2 src16.7 pts$15.00
19Muse Spark 1.3Meta84th6 bench · 2 src3.9 pts$4.25
20DeepSeek V4.1 FlashDeepSeek84th9 bench · 2 src—$1.20
21Grok 4xAI82nd5 bench · 2 src—$15.00
22Muse Spark 1.2Meta82nd3 bench · 2 src—$4.25
23Gemini 3 Flash PreviewGoogle81st10 bench · 2 src4.4 pts$3.00
24GLM-5.3Zhipu80th8 bench · 2 src—$4.40
25Muse Spark 1.1Meta80th3 bench · 2 src—$4.25
26Gemini 3.5 FlashGoogle80th14 bench · 3 src7.5 pts$9.00
27Claude Opus 4.7Anthropic79th14 bench · 2 src11.1 pts$25.00
28Grok 4.6xAI79th10 bench · 2 src0.6 pts$6.00
29DeepSeek R1DeepSeek78th4 bench · 2 src—$2.60
30Kimi K3Kimi78th9 bench · 2 src16.5 pts$1.00
31Gemini 3.1 Pro PreviewGoogle77th13 bench · 2 src0.3 pts$12.00
32Claude Sonnet 5Anthropic76th8 bench · 2 src10.6 pts$10.00
33Claude Haiku 5.5Anthropic75th10 bench · 2 src4.5 pts$0.50
34Kimi K2.5Kimi75th6 bench · 2 src3.0 pts$2.80
35Qwen3.7-MaxAlibaba74th10 bench · 2 src—$7.50
36GPT 5OpenAI73rd13 bench · 3 src8.8 pts$10.00
37DeepSeek-V3.2-ExpDeepSeek71st6 bench · 3 src10.0 pts$0.89
38Grok 4.7xAI71st9 bench · 2 src—$6.00
39Kimi K2 InstructKimi71st6 bench · 3 src—$2.5146
40GPT 5.1OpenAI71st8 bench · 2 src8.3 pts$10.00
41o3OpenAI70th11 bench · 3 src4.2 pts$8.00
42DeepSeek V4 Flash 0731DeepSeek70th8 bench · 2 src—$0.28
43GPT 5.2OpenAI69th13 bench · 3 src14.3 pts$14.00
44GLM 5Zhipu68th5 bench · 2 src0.7 pts$3.20
45Gemini 3.6 FlashGoogle68th9 bench · 2 src10.6 pts$3.75
46Claude Opus 4Anthropic68th7 bench · 2 src4.6 pts—
47Grok 4.5xAI67th8 bench · 2 src—$6.00
48GPT-5.6 LunaOpenAI67th9 bench · 2 src26.0 pts$1.20
49Claude Sonnet 4Anthropic65th7 bench · 3 src11.6 pts—
50Claude Opus 4.6Anthropic65th12 bench · 2 src3.3 pts$25.00
51Claude 3.7 SonnetAnthropic64th7 bench · 3 src12.5 pts$15.00
52Claude Sonnet 4.6Anthropic63rd8 bench · 2 src7.8 pts$15.94
53Qwen3.6 PlusAlibaba62nd9 bench · 2 src—$3.00
54Kimi K2.6Kimi61st13 bench · 2 src—$4.00
55GPT-4.5OpenAI61st4 bench · 2 src—$150.00
56GPT-6 LunaOpenAI60th12 bench · 2 src—$0.50
57GLM-5.2Zhipu59th10 bench · 2 src4.0 pts$4.40
58Qwen3.6 27BAlibaba59th4 bench · 2 src—$2.4754
59Grok 3 Betanot in catalogxAI58th6 bench · 2 src——
60GLM-5.3-FlashZhipu56th7 bench · 2 src—$0.50
61Kimi K2.7 CodeKimi54th7 bench · 2 src—$4.00
62Mistral Large 4Mistral53rd9 bench · 2 src—$4.4156
63InklingThinking Machines53rd7 bench · 2 src—$4.05
64o1-2024-12-17OpenAI52nd9 bench · 3 src—$54.00
65Gemini 2.5 ProGoogle51st7 bench · 2 src4.0 pts$10.00
66o4-miniOpenAI50th12 bench · 3 src—$4.00
67GPT 5 miniOpenAI49th11 bench · 2 src3.3 pts$2.00
68o1-miniOpenAI49th5 bench · 2 src—$4.40
69MiniMax M3MiniMax47th5 bench · 2 src—$1.20
70GPT-5.4 miniOpenAI47th10 bench · 2 src3.3 pts$4.50
71Claude Opus 4.5Anthropic46th12 bench · 2 src4.4 pts$25.00
72GPT-5.4 nanoOpenAI46th10 bench · 2 src18.9 pts$1.25
73Qwen3 32BAlibaba45th3 bench · 2 src——
74gpt-oss-120bOpenAI43rd5 bench · 3 src—free
75QwQ 32BAlibaba43rd3 bench · 2 src—$1.00
76O3 MiniOpenAI39th12 bench · 3 src6.6 pts$4.40
77GLM 4.6Zhipu37th3 bench · 2 src—$2.20
78GPT 4.1OpenAI36th10 bench · 3 src9.0 pts$8.00
79Gemini 2.0 Flash ThinkingGoogle35th6 bench · 3 src—$0.42
80GLM 4.5Zhipu33rd3 bench · 2 src—$2.20
81GPT 5 nanoOpenAI33rd11 bench · 2 src9.5 pts$0.40
82Gemini 3.5 Flash-LiteGoogle32nd7 bench · 2 src11.5 pts$2.50
83Gemma 3 27BGoogle28th4 bench · 2 src—free
84Llama 4 Maverick InstructMeta28th6 bench · 3 src—$0.60
85GPT 4.1 miniOpenAI27th10 bench · 3 src—$1.60
86GPT-4.1 nanoOpenAI19th7 bench · 2 src—$0.40
87Llama 4 Scout InstructMeta19th5 bench · 2 src—$0.30
88GPT 4oOpenAI18th9 bench · 3 src2.0 pts$10.00
89GPT-4o miniOpenAI13th8 bench · 2 src—$0.60
90Claude 3 OpusAnthropic13th6 bench · 2 src—$75.00
91Claude 2not in catalogAnthropic2nd4 bench · 2 src——

Effort range is the median spread between the same model's reasoning-effort settings, in points. It is larger than the disagreement between sources, so a score quoted without its configuration is under-specified.

Evidence counts the benchmarks behind the number and the independent sources behind those. The percentile denominator never changes with the filter.

Open weights uses the catalog flag. Closed is every catalog model without that flag. A scored model with no catalog row is neither.

Paid is a listed output price above zero. Free is a listed output price of zero. An open-weight model can have a paid API.