基准测试浏览器

对比325个AI模型在四个基准测试类别中的表现：推理、代码生成、语言理解和工具使用。评分基于模型能力和综合排名。

模型总数

325

支持推理

148

支持视觉

136

平均评分

基准测试平均分最高的模型

基准测试类别

每个基准测试类别的顶级模型，基于能力支持和综合排名数据评分。

Reasoning

Chain-of-thought reasoning, logic, and mathematical problem solving.

#	模型	提供商	评分
1	GPT-5.4 ProOpenAI	OpenAI	100
2	GPT-5.4OpenAI	OpenAI	100
3	Claude Opus 4.6Anthropic	Anthropic	100
4	Claude Sonnet 4.6Anthropic	Anthropic	100
5	Claude Sonnet 4.5Anthropic	Anthropic	100
6	Grok 4.1 FastxAI	xAI	100
7	Grok 4.20 BetaxAI	xAI	100
8	Gemini 3 Flash PreviewGoogle	Google	100
9	Grok 4 FastxAI	xAI	100
10	Gemini 3.1 Pro PreviewGoogle	Google	100
11	Gemini 3.1 Pro Preview Custom ToolsGoogle	Google	100
12	Qwen3.5 Plus 2026-02-15Alibaba	Alibaba	100
13	Gemini 2.5 ProGoogle	Google	100
14	Gemini 2.5 Flash Lite Preview 09-2025Google	Google	100
15	Gemini 2.5 Pro Preview 06-05Google	Google	100

Code Generation

Code writing, debugging, and software engineering tasks.

#	模型	提供商	评分
1	GPT-5.4 ProOpenAI	OpenAI	90
2	GPT-5.4OpenAI	OpenAI	90
3	GPT-5.4 MiniOpenAI	OpenAI	90
4	Claude Opus 4.6Anthropic	Anthropic	89
5	GPT-5.2 ProOpenAI	OpenAI	89
6	GPT-5.2OpenAI	OpenAI	89
7	GPT-5 ProOpenAI	OpenAI	89
8	o3 Deep ResearchOpenAI	OpenAI	88
9	Claude Opus 4.5Anthropic	Anthropic	87
10	GPT-5OpenAI	OpenAI	87
11	Gemini 3 Flash PreviewGoogle	Google	87
12	Claude Sonnet 4.6Anthropic	Anthropic	86
13	Claude Sonnet 4.5Anthropic	Anthropic	86
14	Grok 4.1 FastxAI	xAI	85
15	o3 ProOpenAI	OpenAI	85

Language Understanding

Text comprehension, multi-modal input, and information retrieval.

#	模型	提供商	评分
1	GPT-5.4 ProOpenAI	OpenAI	84
2	GPT-5.4OpenAI	OpenAI	84
3	GPT-5.4 MiniOpenAI	OpenAI	83
4	GPT-5.2 ProOpenAI	OpenAI	83
5	GPT-5.2OpenAI	OpenAI	83
6	Claude Opus 4.6Anthropic	Anthropic	82
7	GPT-5 ProOpenAI	OpenAI	82
8	o3 Deep ResearchOpenAI	OpenAI	82
9	Claude Opus 4.5Anthropic	Anthropic	81
10	GPT-5OpenAI	OpenAI	81
11	Claude Sonnet 4.6Anthropic	Anthropic	80
12	Claude Sonnet 4.5Anthropic	Anthropic	80
13	Grok 4.1 FastxAI	xAI	79
14	o3 ProOpenAI	OpenAI	79
15	Grok 4.20 BetaxAI	xAI	78

Tool Use

Function calling, structured output, and agentic workflows.

#	模型	提供商	评分
1	GPT-5.4 ProOpenAI	OpenAI	98
2	GPT-5.4OpenAI	OpenAI	98
3	GPT-5.4 MiniOpenAI	OpenAI	98
4	Claude Opus 4.6Anthropic	Anthropic	98
5	GPT-5.2 ProOpenAI	OpenAI	98
6	GPT-5.2OpenAI	OpenAI	98
7	GPT-5 ProOpenAI	OpenAI	98
8	o3 Deep ResearchOpenAI	OpenAI	98
9	Claude Sonnet 4.6Anthropic	Anthropic	98
10	Claude Opus 4.5Anthropic	Anthropic	98
11	Claude Sonnet 4.5Anthropic	Anthropic	98
12	GPT-5OpenAI	OpenAI	98
13	Grok 4.1 FastxAI	xAI	98
14	Grok 4.20 BetaxAI	xAI	98
15	o3 ProOpenAI	OpenAI	98

跨基准测试对比

全部325个模型按四个基准测试类别的平均分排名。

模型	提供商	Code	推理	Language	Tool Use	平均
GPT-5.4 ProOpenAI	OpenAI	90	100	84	98	93
GPT-5.4OpenAI	OpenAI	90	100	84	98	93
GPT-5.4 MiniOpenAI	OpenAI	90	98	83	98	92
Claude Opus 4.6Anthropic	Anthropic	89	100	82	98	92
GPT-5.2 ProOpenAI	OpenAI	89	98	83	98	92
GPT-5.2OpenAI	OpenAI	89	98	83	98	92
GPT-5 ProOpenAI	OpenAI	89	98	82	98	92
o3 Deep ResearchOpenAI	OpenAI	88	98	82	98	92
Claude Sonnet 4.6Anthropic	Anthropic	86	100	80	98	91
Claude Opus 4.5Anthropic	Anthropic	87	98	81	98	91
Claude Sonnet 4.5Anthropic	Anthropic	86	100	80	98	91
GPT-5OpenAI	OpenAI	87	98	81	98	91
Grok 4.1 FastxAI	xAI	85	100	79	98	91
Grok 4.20 BetaxAI	xAI	84	100	78	98	90
o3 ProOpenAI	OpenAI	85	98	79	98	90
Gemini 3 Flash PreviewGoogle	Google	87	100	73	98	90
GPT-5.1OpenAI	OpenAI	83	98	78	98	89
o3OpenAI	OpenAI	83	98	78	98	89
GPT-5.4 NanoOpenAI	OpenAI	83	98	77	98	89
GPT-5.3-CodexOpenAI	OpenAI	83	98	77	98	89
GPT-5.2-CodexOpenAI	OpenAI	83	98	77	98	89
GPT-5.1-Codex-MaxOpenAI	OpenAI	83	98	77	98	89
o4 Mini Deep ResearchOpenAI	OpenAI	83	98	77	98	89
Grok 4 FastxAI	xAI	82	100	76	98	89
o4 Mini HighOpenAI	OpenAI	83	98	77	98	89
Grok 4xAI	xAI	84	95	78	98	89
o4 MiniOpenAI	OpenAI	82	98	77	98	89
Claude Haiku 4.5Anthropic	Anthropic	81	98	76	98	88
Claude Opus 4.1Anthropic	Anthropic	81	98	75	98	88
Gemini 3.1 Pro PreviewGoogle	Google	83	100	70	98	88
Gemini 3.1 Pro Preview Custom ToolsGoogle	Google	83	100	69	98	88
Qwen3.5 Plus 2026-02-15Alibaba	Alibaba	83	100	69	98	88
Gemini 2.5 ProGoogle	Google	83	100	69	98	88
Gemini 2.5 Flash Lite Preview 09-2025Google	Google	82	100	69	98	87
Gemini 2.5 Pro Preview 06-05Google	Google	82	100	69	98	87
MiMo-V2-OmniXiaomi	Xiaomi	83	98	69	98	87
Seed-2.0-LiteByteDance	ByteDance	83	98	69	98	87
Qwen3.5-9BAlibaba	Alibaba	83	98	69	98	87
Seed-2.0-MiniByteDance	ByteDance	83	98	69	98	87
Kimi K2.5Moonshot AI	Moonshot AI	83	98	69	98	87
Seed 1.6 FlashByteDance	ByteDance	83	98	69	98	87
Seed 1.6ByteDance	ByteDance	83	98	69	98	87
GPT-5.1-CodexOpenAI	OpenAI	83	98	69	98	87
GPT-5.1-Codex-MiniOpenAI	OpenAI	83	98	69	98	87
Qwen3 VL 8B ThinkingAlibaba	Alibaba	83	98	69	98	87
Qwen3 VL 30B A3B ThinkingAlibaba	Alibaba	83	98	69	98	87
GPT-5 CodexOpenAI	OpenAI	83	98	69	98	87
GPT-5 MiniOpenAI	OpenAI	78	98	73	98	87
Gemini 2.5 Pro Preview 05-06Google	Google	81	100	68	98	87
Gemini 3.1 Flash Lite PreviewGoogle	Google	81	100	67	98	87
Gemini 2.5 Flash LiteGoogle	Google	80	100	67	98	86
Qwen3.5-FlashAlibaba	Alibaba	79	100	66	98	86
Qwen3.5 397B A17BAlibaba	Alibaba	80	98	67	98	86
Grok Code Fast 1xAI	xAI	83	95	67	98	86
Gemini 2.5 FlashGoogle	Google	79	100	66	98	86
GPT-5 NanoOpenAI	OpenAI	75	98	71	98	86
Qwen3.5-122B-A10BAlibaba	Alibaba	79	98	66	98	85
MiMo-V2-ProXiaomi	Xiaomi	83	100	59	98	85
Qwen3.5-35B-A3BAlibaba	Alibaba	78	98	65	98	85
Qwen3.5-27BAlibaba	Alibaba	78	98	65	98	85
Mistral Small 4Mistral AI	Mistral AI	79	95	66	98	85
Claude Opus 4Anthropic	Anthropic	75	98	75	90	85
Nemotron 3 Super (free)NVIDIA	NVIDIA	82	98	59	98	84
Qwen3 VL 235B A22B ThinkingAlibaba	Alibaba	77	98	64	98	84
Qwen Plus 0728 (thinking)Alibaba	Alibaba	81	100	58	98	84
MiniMax M2.5 (free)MiniMax	MiniMax	82	98	58	98	84
Claude Sonnet 4Anthropic	Anthropic	74	98	74	90	84
MiniMax M2.7MiniMax	MiniMax	81	98	58	98	84
MiMo-V2-FlashXiaomi	Xiaomi	81	98	58	98	84
Trinity Miniarcee-ai	arcee-ai	81	98	58	98	84
Tongyi DeepResearch 30B A3BAlibaba	Alibaba	81	98	57	98	84
o1OpenAI	OpenAI	75	98	63	98	84
Mercury 2Inception	Inception	80	98	57	98	83
Qwen3 Max ThinkingAlibaba	Alibaba	80	98	57	98	83
gpt-oss-safeguard-20bOpenAI	OpenAI	80	98	57	98	83
Qwen3 30B A3B Thinking 2507Alibaba	Alibaba	80	98	57	98	83
Nemotron Nano 12B 2 VL (free)NVIDIA	NVIDIA	76	98	68	90	83
Claude 3.7 SonnetAnthropic	Anthropic	72	98	72	90	83
Grok 3 MinixAI	xAI	76	95	61	98	83
DeepSeek V3.2 ExpDeepSeek	DeepSeek	77	98	54	98	82
R1 0528DeepSeek	DeepSeek	77	98	54	98	82
MiniMax M2.5MiniMax	MiniMax	76	98	53	98	81
Claude 3.7 Sonnet (thinking)Anthropic	Anthropic	66	98	67	90	80
o3 MiniOpenAI	OpenAI	74	98	51	98	80
MiniMax M2MiniMax	MiniMax	73	98	51	98	80
Qwen3 Next 80B A3B ThinkingAlibaba	Alibaba	73	98	51	98	80
Nova 2 LiteAmazon	Amazon	68	100	61	90	80
DeepSeek V3.2DeepSeek	DeepSeek	74	95	52	98	80
DeepSeek V3.1 TerminusDeepSeek	DeepSeek	74	95	52	98	80
Nemotron 3 SuperNVIDIA	NVIDIA	74	95	51	98	80
Nemotron 3 Nano 30B A3BNVIDIA	NVIDIA	74	95	51	98	80
Solar Pro 3Upstage	Upstage	73	95	51	98	79
MiniMax M2.1MiniMax	MiniMax	73	95	51	98	79
Trinity Mini (free)arcee-ai	arcee-ai	73	95	51	98	79
Kimi K2 ThinkingMoonshot AI	Moonshot AI	73	95	51	98	79
DeepSeek V3 0324DeepSeek	DeepSeek	73	95	51	98	79
Step 3.5 Flash (free)StepFun	StepFun	73	98	55	90	79
Nemotron Nano 9B V2 (free)NVIDIA	NVIDIA	72	95	50	98	79
Nemotron Nano 9B V2NVIDIA	NVIDIA	72	95	50	98	79
Qwen3 235B A22BAlibaba	Alibaba	72	95	50	98	79
Grok 3 Mini BetaxAI	xAI	68	95	54	98	79
DeepSeek V3.1DeepSeek	DeepSeek	74	90	52	98	79
gpt-oss-20bOpenAI	OpenAI	70	98	48	98	79
Qwen3 235B A22B Instruct 2507Alibaba	Alibaba	71	95	49	98	78
Qwen3 30B A3BAlibaba	Alibaba	72	93	50	98	78
Qwen3 14BAlibaba	Alibaba	72	93	50	98	78
Qwen3 32BAlibaba	Alibaba	72	93	50	98	78
ERNIE 4.5 VL 28B A3BBaidu	Baidu	70	90	62	90	78
Composer 2Cursor	Cursor	71	98	53	90	78
Composer 2 FastCursor	Cursor	71	98	53	90	78
Llama 3.3 Nemotron Super 49B V1.5NVIDIA	NVIDIA	70	95	48	98	78
Qwen3 235B A22B Thinking 2507Alibaba	Alibaba	70	95	48	98	78
gpt-oss-120b (free)OpenAI	OpenAI	69	98	52	90	77
gpt-oss-120bOpenAI	OpenAI	69	95	47	98	77
gpt-oss-20b (free)OpenAI	OpenAI	69	98	52	90	77
o3 Mini HighOpenAI	OpenAI	67	98	46	98	77
Step 3.5 FlashStepFun	StepFun	69	98	51	90	77
MiniMax M1MiniMax	MiniMax	65	100	48	90	76
Grok 4.20 Multi-Agent BetaxAI	xAI	81	100	76	43	75
Qwen3 8BAlibaba	Alibaba	67	90	45	98	75
Sonar Pro SearchPerplexity	Perplexity	83	95	77	43	75
Qwen3 4B (free)Alibaba	Alibaba	65	90	44	98	74
GPT-5.3 ChatOpenAI	OpenAI	73	48	77	98	74
Nemotron 3 Nano 30B A3B (free)NVIDIA	NVIDIA	64	95	47	90	74
GPT-5.1 ChatOpenAI	OpenAI	73	48	77	98	74
R1DeepSeek	DeepSeek	65	93	48	90	74
GPT-5 Image MiniOpenAI	OpenAI	50	98	48	98	74
GPT-5 ImageOpenAI	OpenAI	50	98	48	98	74
GPT-4.1 NanoOpenAI	OpenAI	69	53	74	98	74
GPT-5.2 ChatOpenAI	OpenAI	71	48	76	98	73
GPT-4.1OpenAI	OpenAI	67	53	72	98	73
GPT-4.1 MiniOpenAI	OpenAI	67	53	72	98	73
Qwen3 VL 32B InstructAlibaba	Alibaba	70	48	67	98	71
Qwen3 VL 8B InstructAlibaba	Alibaba	70	48	67	98	71
Qwen3 VL 30B A3B InstructAlibaba	Alibaba	70	48	67	98	71
Llama 4 MaverickMeta	Meta	66	53	64	98	70
o1-proOpenAI	OpenAI	76	98	63	43	70
Gemini 2.0 Flash LiteGoogle	Google	65	50	63	98	69
Gemini 2.0 FlashGoogle	Google	65	50	62	98	69
Qwen3 Coder PlusAlibaba	Alibaba	68	53	55	98	69
Qwen3 Coder FlashAlibaba	Alibaba	67	53	55	98	68
Reka Edgereka	reka	67	43	64	98	68
DeepSeek V3.2 SpecialeDeepSeek	DeepSeek	77	98	54	43	68
Nemotron Nano 12B 2 VLNVIDIA	NVIDIA	73	95	61	43	68
Qwen Plus 0728Alibaba	Alibaba	66	53	54	98	68
Nova Premier 1.0Amazon	Amazon	62	53	64	90	67
Qwen3 VL 235B A22B InstructAlibaba	Alibaba	64	45	62	98	67
Llama 4 ScoutMeta	Meta	63	48	60	98	67
QwQ 32BAlibaba	Alibaba	48	98	33	90	67
Ministral 3 14B 2512Mistral AI	Mistral AI	64	45	61	98	67
Ministral 3 8B 2512Mistral AI	Mistral AI	64	45	61	98	67
Mistral Large 3 2512Mistral AI	Mistral AI	64	45	61	98	67
Ministral 3 3B 2512Mistral AI	Mistral AI	63	45	61	98	67
KAT-Coder-Pro V1Kuaishou	Kuaishou	67	48	54	98	67
Grok 3xAI	xAI	64	45	60	98	67
Qwen3 Coder NextAlibaba	Alibaba	66	48	54	98	67
Trinity Large Preview (free)arcee-ai	arcee-ai	63	45	59	98	66
Qwen3 MaxAlibaba	Alibaba	66	48	53	98	66
Qwen3 30B A3B Instruct 2507Alibaba	Alibaba	65	48	53	98	66
Hunyuan A13B InstructTencent	Tencent	73	98	50	43	66
Mistral Medium 3.1Mistral AI	Mistral AI	61	45	59	98	66
Qwen VL MaxAlibaba	Alibaba	59	48	58	98	66
SWE-1.5Windsurf	Windsurf	49	90	34	90	66
LongCat Flash ChatMeituan	Meituan	63	48	51	98	65
Qwen3 Coder 30B A3B InstructAlibaba	Alibaba	63	48	51	98	65
Kimi K2 0711Moonshot AI	Moonshot AI	63	48	51	98	65
GPT-4o AudioOpenAI	OpenAI	63	48	50	98	65
Mistral Small 3.2 24BMistral AI	Mistral AI	59	45	57	98	65
GPT-4o-miniOpenAI	OpenAI	57	48	55	98	65
GPT-4oOpenAI	OpenAI	56	48	55	98	64
DeepSeek V3DeepSeek	DeepSeek	61	48	49	98	64
Jamba Large 1.7AI21 Labs	AI21 Labs	62	45	50	98	64
Mistral Medium 3Mistral AI	Mistral AI	57	45	55	98	64
GPT-4o (2024-11-20)OpenAI	OpenAI	55	48	54	98	64
Cogito v2.1 671Bdeepcogito	deepcogito	68	95	47	43	63
Qwen3 Next 80B A3B InstructAlibaba	Alibaba	61	45	49	98	63
Qwen-PlusAlibaba	Alibaba	57	53	45	98	63
ERNIE 4.5 VL 424B A47B Baidu	Baidu	65	93	59	35	63
Mercury CoderInception	Inception	58	48	47	98	63
Grok 3 BetaxAI	xAI	56	45	52	98	63
Mistral Small 3.1 24B (free)Mistral AI	Mistral AI	55	45	53	98	63
Olmo 3 32B ThinkAllen AI	Allen AI	68	93	46	43	63
Sonar Reasoning ProPerplexity	Perplexity	59	95	61	35	63
Devstral 2 2512Mistral AI	Mistral AI	59	45	47	98	62
Qwen3 Next 80B A3B Instruct (free)Alibaba	Alibaba	59	45	47	98	62
Llama 3.3 70B InstructMeta	Meta	57	48	46	98	62
Olmo 3.1 32B ThinkAllen AI	Allen AI	67	93	45	43	62
ERNIE 4.5 21B A3B ThinkingBaidu	Baidu	66	98	49	35	62
R1 Distill Llama 70BDeepSeek	DeepSeek	64	98	43	43	62
GPT-4 TurboOpenAI	OpenAI	53	45	52	98	62
Claude 3.5 HaikuAnthropic	Anthropic	50	45	62	90	62
Aion-2.0aion-labs	aion-labs	65	98	48	35	62
Kimi K2 0905Moonshot AI	Moonshot AI	57	45	46	98	62
MercuryInception	Inception	56	48	44	98	62
Codestral 2508Mistral AI	Mistral AI	57	45	45	98	61
Qwen3 Coder 480B A35BAlibaba	Alibaba	56	45	45	98	61
Claude 3.5 SonnetAnthropic	Anthropic	53	45	56	90	61
GPT-4o (2024-08-06)OpenAI	OpenAI	49	48	49	98	61
Devstral MediumMistral AI	Mistral AI	55	45	44	98	61
Devstral Small 1.1Mistral AI	Mistral AI	55	45	44	98	61
GPT-4o-mini (2024-07-18)OpenAI	OpenAI	48	48	48	98	61
GPT-4o (extended)OpenAI	OpenAI	48	48	48	98	61
Qwen3 Coder 480B A35B (free)Alibaba	Alibaba	55	48	48	90	60
R1 Distill Qwen 32BDeepSeek	DeepSeek	63	93	42	43	60
Pixtral Large 2411Mistral AI	Mistral AI	49	45	49	98	60
Olmo 3.1 32B InstructAllen AI	Allen AI	57	40	45	98	60
Rnj 1 Instructessentialai	essentialai	57	40	45	98	60
Llama 3.1 Nemotron Ultra 253B v1NVIDIA	NVIDIA	61	95	40	43	60
Qwen-TurboAlibaba	Alibaba	53	45	42	98	60
Llama 3.1 70B InstructMeta	Meta	53	45	42	98	60
GPT-4o (2024-05-13)OpenAI	OpenAI	47	45	47	98	59
Mistral Small 3Mistral AI	Mistral AI	52	43	42	98	59
Nova Lite 1.0Amazon	Amazon	46	45	51	90	58
Nova Pro 1.0Amazon	Amazon	46	45	51	90	58
Virtuoso Largearcee-ai	arcee-ai	50	48	43	90	58
Sonar Deep ResearchPerplexity	Perplexity	54	95	47	35	58
Qwen-Max Alibaba	Alibaba	52	40	41	98	58
Llama 3.1 Nemotron 70B InstructNVIDIA	NVIDIA	47	48	37	98	58
Mistral LargeMistral AI	Mistral AI	48	45	38	98	57
ERNIE 4.5 21B A3BBaidu	Baidu	52	40	46	90	57
Aion-1.0aion-labs	aion-labs	55	98	39	35	57
Aion-1.0-Miniaion-labs	aion-labs	55	98	39	35	57
Mistral Large 2407Mistral AI	Mistral AI	47	45	37	98	57
Nano Banana 2 (Gemini 3.1 Flash Image Preview)Google	Google	50	93	40	43	57
Nano Banana Pro (Gemini 3 Pro Image Preview)Google	Google	50	93	40	43	57
GPT-5 ChatOpenAI	OpenAI	65	48	70	43	57
Mistral NemoMistral AI	Mistral AI	45	48	35	98	57
Qwen2.5 72B InstructAlibaba	Alibaba	47	43	37	98	56
LFM2.5-1.2B-Thinking (free)Liquid AI	Liquid AI	57	90	41	35	56
Mistral Large 2411Mistral AI	Mistral AI	45	45	35	98	56
SabaMistral AI	Mistral AI	47	40	37	98	56
Command R+ (08-2024)Cohere	Cohere	43	45	33	98	55
Command R (08-2024)Cohere	Cohere	43	45	33	98	55
Mistral Small CreativeMistral AI	Mistral AI	47	40	41	90	55
Qwen2.5 7B InstructAlibaba	Alibaba	41	43	32	98	54
GPT-4o-mini Search PreviewOpenAI	OpenAI	63	48	59	43	53
Nova Micro 1.0Amazon	Amazon	41	45	36	90	53
GPT-4 Turbo PreviewOpenAI	OpenAI	39	45	30	98	53
GPT-4 Turbo (older v1106)OpenAI	OpenAI	39	45	30	98	53
Llama 3.1 8B InstructMeta	Meta	39	43	30	98	53
Mixtral 8x7B InstructMistral AI	Mistral AI	39	43	30	98	53
Claude 3 HaikuAnthropic	Anthropic	34	45	40	90	52
Voxtral Small 24B 2507Mistral AI	Mistral AI	40	40	30	98	52
Mistral Small 3.1 24BMistral AI	Mistral AI	58	48	56	43	51
GPT-3.5 Turbo 16kOpenAI	OpenAI	37	40	28	98	51
GPT-3.5 TurboOpenAI	OpenAI	37	40	28	98	51
Gemma 3 27BGoogle	Google	56	48	54	43	50
GPT-4 (older v0314)OpenAI	OpenAI	36	40	27	98	50
GPT-4OpenAI	OpenAI	36	40	27	98	50
GPT-3.5 Turbo (older v0613)OpenAI	OpenAI	35	40	27	98	50
GPT AudioOpenAI	OpenAI	60	48	48	43	50
GPT Audio MiniOpenAI	OpenAI	60	48	48	43	50
GPT-4o Search PreviewOpenAI	OpenAI	56	48	52	43	50
Mixtral 8x22B InstructMistral AI	Mistral AI	35	40	26	98	50
Gemma 3 27B (free)Google	Google	55	45	54	43	49
Llama 3.3 70B Instruct (free)Meta	Meta	35	40	31	90	49
Qwen VL PlusAlibaba	Alibaba	54	45	53	43	49
Sonar ProPerplexity	Perplexity	50	45	62	35	48
Llama Guard 4 12BMeta	Meta	52	45	51	43	48
Qwen2.5 VL 72B InstructAlibaba	Alibaba	53	43	52	43	48
Llama 3 8B InstructMeta	Meta	29	43	21	98	48
Gemma 3 4B (free)Google	Google	54	40	53	43	48
MiniMax-01MiniMax	MiniMax	49	53	53	35	48
Qwen2.5 VL 32B InstructAlibaba	Alibaba	50	45	50	43	47
Gemma 3 4BGoogle	Google	50	45	49	43	47
Gemma 3 12BGoogle	Google	50	45	49	43	47
Llama 3.2 11B Vision InstructMeta	Meta	48	48	48	43	47
Spotlightarcee-ai	arcee-ai	50	48	53	35	47
UI-TARS 7B ByteDance	ByteDance	50	45	54	35	46
ERNIE 4.5 300B A47B Baidu	Baidu	56	40	44	43	46
Command ACohere	Cohere	53	45	42	43	46
Palmyra X5Writer	Writer	52	50	45	35	46
Phi 4Microsoft	Microsoft	53	43	42	43	45
Gemma 2 27BGoogle	Google	53	40	42	43	45
Gemma 3n 2B (free)Google	Google	51	40	41	43	44
SonarPerplexity	Perplexity	43	40	55	35	43
Gemma 3n 4B (free)Google	Google	49	40	39	43	43
Gemma 3 12B (free)Google	Google	44	40	49	35	42
Nano Banana (Gemini 2.5 Flash Image)Google	Google	40	43	40	43	42
Maestro Reasoningarcee-ai	arcee-ai	44	48	39	35	42
MiniMax M2-herMiniMax	MiniMax	48	40	42	35	41
Granite 4.0 MicroIBM	IBM	44	45	39	35	41
Command R7B (12-2024)Cohere	Cohere	41	45	31	43	40
LFM2-24B-A2BLiquid AI	Liquid AI	43	40	37	35	39
LFM2.5-1.2B-Instruct (free)Liquid AI	Liquid AI	43	40	37	35	39
LFM2-8B-A1BLiquid AI	Liquid AI	43	40	37	35	39
LFM2-2.6BLiquid AI	Liquid AI	43	40	37	35	39
Qwen2.5 Coder 7B InstructAlibaba	Alibaba	39	40	30	43	38
Llama 3 70B InstructMeta	Meta	37	40	28	43	37
Llemma 7beleutherai	eleutherai	38	40	33	35	37
Olmo 2 32B InstructAllen AI	Allen AI	35	45	31	35	37
Gemma 3n 4BGoogle	Google	37	40	32	35	36
Llama Guard 3 8BMeta	Meta	34	45	30	35	36
Coder Largearcee-ai	arcee-ai	36	40	32	35	36
Qwen2.5 Coder 32B InstructAlibaba	Alibaba	34	40	30	35	35
GPT-3.5 Turbo InstructOpenAI	OpenAI	31	40	23	43	34
Midjourney v6.1Midjourney	Midjourney	35	40	30	30	34
DALL-E 3OpenAI	OpenAI	35	40	30	30	34
Stable Diffusion 3.5Stability AI	Stability AI	35	40	30	30	34
FLUX.1 ProBlack Forest Labs	Black Forest Labs	35	40	30	30	34
Ideogram 2.0Ideogram	Ideogram	35	40	30	30	34
Recraft V3Recraft	Recraft	35	40	30	30	34
Imagen 3Google	Google	35	40	30	30	34
Adobe Firefly 3Adobe	Adobe	35	40	30	30	34
Leonardo PhoenixLeonardo AI	Leonardo AI	35	40	30	30	34
SoraOpenAI	OpenAI	35	40	30	30	34
Veo 2Google	Google	35	40	30	30	34
Runway Gen-3 AlphaRunway	Runway	35	40	30	30	34
Kling 1.6Kuaishou	Kuaishou	35	40	30	30	34
Pika 2.0Pika	Pika	35	40	30	30	34
MiniMax Video-01MiniMax	MiniMax	35	40	30	30	34
Luma Dream MachineLuma AI	Luma AI	35	40	30	30	34
Stable Video DiffusionStability AI	Stability AI	35	40	30	30	34
Wan 2.1 T2VWan AI	Wan AI	35	40	30	30	34
LTX-Video 2Lightricks	Lightricks	35	40	30	30	34
Llama 3.2 3B Instruct (free)Meta	Meta	28	45	25	35	33
autofixer-01Vercel	Vercel	31	40	27	35	33
Inflection 3 ProductivityInflection	Inflection	29	40	26	35	33
Inflection 3 PiInflection	Inflection	29	40	26	35	33
Llama 3.2 3B InstructMeta	Meta	29	40	25	35	32
MellumJetBrains	JetBrains	26	40	23	35	31
WizardLM-2 8x22BMicrosoft	Microsoft	26	40	22	35	31
Llama 3.2 1B InstructMeta	Meta	25	40	22	35	31
Gemma 2 9BGoogle	Google	24	40	21	35	30
Mistral 7B Instruct v0.1Mistral AI	Mistral AI	16	40	14	35	26

探索更多

通过功能筛选、价格对比或完整排行榜深入探索。

全部探索器功能基准测试概览排行榜对比模型

Frequently Asked Questions

AI benchmarks are standardized tests that measure model performance on specific tasks - math (MATH-500), coding (HumanEval, SWE-bench), reasoning (GPQA), knowledge (MMLU), and conversational ability (Arena ELO). They help compare models objectively.

No single benchmark tells the full story. Arena ELO (crowdsourced human preferences) best reflects real-world quality. SWE-bench is most relevant for coding. MATH-500 and GPQA test reasoning depth. Use multiple benchmarks for evaluation.

Benchmarks have limitations - models can be trained on benchmark data (contamination), and high scores don't always predict real-world performance. Arena ELO is hardest to game since it uses blind human comparisons. Use benchmarks as one signal among many.