最佳角色扮演AI模型

使用Arena Elo评分(归一化到0-100刻度)对AI模型的角色扮演和创意对话质量进行排名。

Last updated: just now

第一名模型

Claude Opus 4.6

评分: 100

平均评分

76.6

所有排名模型

已排名模型

有基准测试数据

权重：Arena Elo (100%)

#	模型	提供商	评分	Arena Elo
1	Claude Opus 4.6Anthropic	Anthropic	100	100
2	Grok 4.20 BetaxAI	xAI	99.3	99.3
3	Gemini 3.1 Pro PreviewGoogle	Google	98.7	98.7
4	GPT-5.4OpenAI	OpenAI	97.5	97.5
5	GPT-5.2 ChatOpenAI	OpenAI	96.8	96.8
6	GPT-5.2OpenAI	OpenAI	96.8	96.8
7	Gemini 3 Flash PreviewGoogle	Google	95.7	95.7
8	Grok 4.1 FastxAI	xAI	95.5	95.5
9	GPT-5OpenAI	OpenAI	94.2	94.2
10	Grok 4xAI	xAI	93.7	93.7
11	Claude Sonnet 4.6Anthropic	Anthropic	93.3	93.3
12	GPT-5.1OpenAI	OpenAI	92.7	92.7
13	Claude Sonnet 4.5Anthropic	Anthropic	92	92
14	Qwen3.5 397B A17BAlibaba	Alibaba	91.7	91.7
15	Claude Opus 4.1Anthropic	Anthropic	91.5	91.5
16	Gemini 2.5 ProGoogle	Google	90.7	90.7
17	Gemini 3.1 Flash Lite PreviewGoogle	Google	89.5	89.5
18	Claude Opus 4.5Anthropic	Anthropic	88.3	88.3
19	GPT-5 ChatOpenAI	OpenAI	87.8	87.8
20	DeepSeek V3.2 ExpDeepSeek	DeepSeek	87.3	87.3
21	DeepSeek V3.2DeepSeek	DeepSeek	87.2	87.2
22	Grok 4 FastxAI	xAI	87	87
23	Claude Opus 4Anthropic	Anthropic	86.7	86.7
24	Qwen3.5-122B-A10BAlibaba	Alibaba	86.5	86.5
25	DeepSeek V3.1DeepSeek	DeepSeek	86.5	86.5
26	DeepSeek V3.1 TerminusDeepSeek	DeepSeek	86.2	86.2
27	Qwen3 VL 235B A22B InstructAlibaba	Alibaba	86	86
28	o3OpenAI	OpenAI	85.8	85.8
29	Qwen3.5-27BAlibaba	Alibaba	85	85
30	MiniMax M2.5MiniMax	MiniMax	84	84
31	Qwen3 Next 80B A3B InstructAlibaba	Alibaba	83.7	83.7
32	LongCat Flash ChatMeituan	Meituan	83.5	83.5
33	Qwen3.5-FlashAlibaba	Alibaba	83.3	83.3
34	Qwen3.5-35B-A3BAlibaba	Alibaba	83	83
35	Qwen3 VL 235B A22B ThinkingAlibaba	Alibaba	82.7	82.7
36	Gemini 2.5 FlashGoogle	Google	82.5	82.5
37	GPT-5 MiniOpenAI	OpenAI	81.8	81.8
38	Step 3.5 FlashStepFun	StepFun	81.5	81.5
39	Claude 3.7 Sonnet (thinking)Anthropic	Anthropic	81.3	81.3
40	Claude Sonnet 4Anthropic	Anthropic	81.2	81.2
41	R1 0528DeepSeek	DeepSeek	80	80
42	o4 MiniOpenAI	OpenAI	80	80
43	o3 MiniOpenAI	OpenAI	78.5	78.5
44	R1DeepSeek	DeepSeek	78.2	78.2
45	Qwen3 Next 80B A3B ThinkingAlibaba	Alibaba	78.2	78.2
46	MiniMax M1MiniMax	MiniMax	77.8	77.8
47	o3 Mini HighOpenAI	OpenAI	77.3	77.3
48	Grok 3 Mini BetaxAI	xAI	76.3	76.3
49	gpt-oss-120bOpenAI	OpenAI	75.8	75.8
50	Claude 3.7 SonnetAnthropic	Anthropic	75.7	75.7
51	Command ACohere	Cohere	75.7	75.7
52	Gemini 2.0 FlashGoogle	Google	75.3	75.3
53	o1OpenAI	OpenAI	75	75
54	MiniMax M2MiniMax	MiniMax	74.7	74.7
55	Qwen3 8BAlibaba	Alibaba	74.7	74.7
56	GPT-4o (2024-05-13)OpenAI	OpenAI	74.3	74.3
57	DeepSeek V3 0324DeepSeek	DeepSeek	74.2	74.2
58	Grok 3xAI	xAI	73.7	73.7
59	Llama 3.3 Nemotron Super 49B V1.5NVIDIA	NVIDIA	73.7	73.7
60	GPT-5 NanoOpenAI	OpenAI	73	73
61	Nova 2 LiteAmazon	Amazon	73	73
62	QwQ 32BAlibaba	Alibaba	72.7	72.7
63	GPT-4o (2024-08-06)OpenAI	OpenAI	72.5	72.5
64	Olmo 3.1 32B InstructAllen AI	Allen AI	71.8	71.8
65	Llama 4 MaverickMeta	Meta	70.8	70.8
66	DeepSeek V3DeepSeek	DeepSeek	69.7	69.7
67	GPT-4o-mini (2024-07-18)OpenAI	OpenAI	69.7	69.7
68	gpt-oss-20bOpenAI	OpenAI	69.7	69.7
69	Mistral Large 2407Mistral AI	Mistral AI	69	69
70	MercuryInception	Inception	68.2	68.2
71	Olmo 3 32B ThinkAllen AI	Allen AI	67.8	67.8
72	Qwen2.5 72B InstructAlibaba	Alibaba	67.2	67.2
73	GPT-4.1OpenAI	OpenAI	66.7	66.7
74	Llama 3.1 Nemotron 70B InstructNVIDIA	NVIDIA	66.5	66.5
75	GPT-4oOpenAI	OpenAI	64.3	64.3
76	Olmo 3.1 32B ThinkAllen AI	Allen AI	64.2	64.2
77	Mistral LargeMistral AI	Mistral AI	63.3	63.3
78	Llama 3 70B InstructMeta	Meta	62.7	62.7
79	Qwen2.5 Coder 32B InstructAlibaba	Alibaba	61.8	61.8
80	Claude 3.5 SonnetAnthropic	Anthropic	61.8	61.8
81	Claude 3 HaikuAnthropic	Anthropic	60.2	60.2
82	Phi 4Microsoft	Microsoft	59.3	59.3
83	GPT-4 TurboOpenAI	OpenAI	59.2	59.2
84	Command R (08-2024)Cohere	Cohere	58.3	58.3
85	Llama 3.3 70B InstructMeta	Meta	57.2	57.2
86	Claude Haiku 4.5Anthropic	Anthropic	56.7	56.7
87	Llama 3 8B InstructMeta	Meta	53.8	53.8
88	GPT-4o-miniOpenAI	OpenAI	53.7	53.7
89	Llama 3.1 8B InstructMeta	Meta	52	52
90	Llama 3.1 70B InstructMeta	Meta	49.7	49.7
91	Claude 3.5 HaikuAnthropic	Anthropic	46.3	46.3
92	Llama 3.2 3B InstructMeta	Meta	44.5	44.5
93	Llama 3.2 1B InstructMeta	Meta	35.2	35.2

评分计算方式

每个模型的评分是其可用基准测试结果的加权平均值。当模型缺少某些基准测试时，权重会在可用的基准测试之间重新归一化。所有评分均为0-100的刻度。数据来源于官方模型卡片、已发表论文和第三方评估平台。

其他专业排行榜

编程最佳数学最佳推理最佳写作最佳指令遵循最佳数据分析最佳多语言最佳

Frequently Asked Questions

根据我们的基准测试分析，Anthropic的Claude Opus 4.6目前在Roleplay领域排名第一，加权评分为100/100。

模型使用Arena Elo基准测试分数的加权平均进行排名。所有分数均归一化到0-100的刻度。

我们目前对93个拥有相关基准测试数据的模型进行了Roleplay任务排名。