中文大模型能力榜

📊 全部数据 · 总览

按 LMArena 中文竞技场(Chinese Arena)的真人中文盲测对战评分,排出大模型在中文场景的能力高下。它只统计中文对话里的偏好胜负,和上方「总榜」(以英文为主)可能明显不同——国产模型在这里通常更靠前。评分反映偏好、非绝对能力,头部分差常落在置信区间内(见下表),宜看作近似并列。数据归因 LMArena,不构成投资或采购建议。榜单发布于 2026-10-15

排名	模型	机构	Arena 分	置信区间	投票数
1	Gemini 4 Argon (High)	google	1608	±31	363
2	Claude Fable 5.1 (Max)	anthropic	1586	±18	1.1K
3	Claude Opus 5.5 (High)	anthropic	1584	±22	767
4	Claude Opus 5 (Max)	anthropic	1573	±13	2.6K
5	Claude Opus 5 (High)	anthropic	1568	±9	5.3K
6	Gemini 3.7 Flash (High)	google	1551	±14	2.1K
7	Gemini 3.8 Flash (High)	google	1551	±12	3.1K
8	Claude Opus 4.6 (High)	anthropic	1550	±9	5.6K
9	Claude Opus 4.6	anthropic	1548	±9	5.9K
10	Claude Fable 5 (High)	anthropic	1542	±11	3.1K
11	Muse Wonder (Max)	meta	1540	±36	248
12	claude-sonnet-5.5-xhigh	anthropic	1537	±25	553
13	qwen3.8-max	alibaba	1536	±14	2.1K
14	gpt-5.5	openai	1533	±10	4.5K
15	qwen3.5-max-preview	alibaba	1533	±15	1.6K
16	Claude Opus 4.7 (High)	anthropic	1532	±10	4.2K
17	Claude Opus 4.7	anthropic	1531	±10	4.1K
18	gemini-3.6-flash-high	google	1530	±12	2.7K
19	qwen3.7-max-preview	alibaba	1530	±34	317
20	glm-5.3-max	zai	1529	±15	1.7K
21	Muse Spark 1.3 (Max)	meta	1528	±16	1.4K
22	Kimi K3 (Max)	moonshot	1528	±13	2.4K
23	mimo-v2.6-pro	xiaomi	1528	±33	299
24	glm-5.3-flash	zai	1528	±12	2.5K
25	Gemini 3.1 Pro Preview	google	1528	±7	8.8K
26	gpt-5.6-sol-xhigh	openai	1527	±11	3.1K
27	gemini-3.5-flash-medium	google	1526	±11	3.4K
28	gemini-3.5-flash-high	google	1525	±11	3.6K
29	gpt-5.5-high	openai	1525	±10	4.5K
30	Gemini 3 Pro	google	1523	±10	3.5K
31	kimi-k2.6	moonshot	1521	±13	2.3K
32	glm-5.2-max	zai	1520	±10	3.6K
33	gpt-5.4-high	openai	1518	±10	4.1K
34	Muse Spark 1.1	meta	1516	±11	3.4K
35	gpt-5.6-terra-xhigh	openai	1515	±11	3.4K
36	glm-5.1	zai	1514	±10	4.4K
37	mimo-v2.6-flash	xiaomi	1513	±20	871
38	ernie-5.0-preview-1022	baidu	1512	±30	390
39	gemini-3-flash	google	1511	±12	2.5K
40	glm-5	zai	1511	±13	2.0K
41	qwen3.7-plus	alibaba	1510	±12	2.7K
42	Muse Spark 1.2 (xHigh)	meta	1509	±23	717
43	gpt-5.4	openai	1508	±10	4.4K
44	Step 5 Preview	stepfun	1508	±24	573
45	Muse Spark	meta	1508	±20	976
46	gemini-2.5-pro	google	1507	±7	9.3K
47	ernie-5.1	baidu	1507	±13	2.1K
48	mimo-v2.5-pro	xiaomi	1507	±9	4.9K
49	claude-opus-4-8	anthropic	1507	±10	4.4K
50	claude-opus-4-8-high	anthropic	1507	±10	4.4K
51	deepseek-v4.1-flash-max	deepseek	1503	±19	973
52	qwen3.8-27b	alibaba	1502	±14	1.7K
53	ernie-5.0-preview-1203	baidu	1501	±21	830
54	glm-4.6	zai	1499	±11	2.7K
55	qwen3.5-397b-a17b	alibaba	1499	±8	6.5K
56	nvidia-nemotron-3-ultra-550b-a55b-nvfp4	nvidia	1497	±22	727
57	gemma-4-26b-a4b	google	1495	±25	524
58	glm-4.7	zai	1495	±18	1.0K
59	gpt-5.1-high	openai	1495	±10	3.4K
60	grok-4.5	xai	1495	±11	3.1K
61	gpt-6.1-sol-max	openai	1495	±25	575
62	kimi-k2.5-thinking	moonshot	1495	±9	4.6K
63	gpt-6-astra-max	openai	1493	±19	917
64	ernie-5.0-0110	baidu	1492	±12	2.5K
65	mistral-large-4	mistral	1492	±23	658
66	inkling	thinky	1492	±11	3.2K
67	claude-sonnet-4-6	anthropic	1492	±9	4.5K
68	glm-5v-turbo	zai	1491	±21	769
69	dola-seed-2.0-pro	bytedance	1489	±9	5.3K
70	qwen3.6-max-preview	alibaba	1488	±31	357
71	hy3	tencent	1487	±20	928
72	deepseek-v4-pro-high-preview	deepseek	1486	±11	3.3K
73	qwen3-235b-a22b-thinking-2507	alibaba	1481	±26	536
74	deepseek-v4-pro	deepseek	1481	±11	3.4K
75	grok-4.20-beta-0309-reasoning	xai	1480	±10	4.2K
76	deepseek-v4-pro-high-20260813	deepseek	1480	±22	705
77	qwen3-max-preview	alibaba	1478	±13	2.1K
78	qwen3.5-27b	alibaba	1478	±13	2.1K
79	claude-sonnet-5-high	anthropic	1478	±10	3.8K
80	gemini-3-flash (thinking-minimal)	google	1477	±8	6.0K
81	grok-4.20-beta1	xai	1476	±15	1.7K
82	qwen3.6-plus	alibaba	1476	±12	2.7K
83	gemma-4-31b	google	1475	±25	522
84	grok-4.6-high	xai	1475	±13	2.3K
85	grok-4.20-multi-agent-beta-0309	xai	1475	±10	4.0K
86	claude-opus-4-5-20251101	anthropic	1470	±8	5.4K
87	gpt-5.6-luna-xhigh	openai	1469	±11	3.2K
88	deepseek-v3.1-thinking	deepseek	1469	±18	1.1K
89	mimo-v2.5	xiaomi	1468	±12	2.6K
90	deepseek-v4-flash-high-preview	deepseek	1468	±11	3.0K
91	mimo-v2-omni	xiaomi	1468	±19	1.0K
92	minimax-m3	minimax	1467	±10	4.1K
93	gemini-3.5-flash-lite	google	1467	±12	2.8K
94	claude-opus-4-5-20251101-high-32k	anthropic	1467	±11	3.0K
95	gpt-5.1	openai	1466	±10	3.6K
96	glm-4.5	zai	1465	±14	1.8K
97	qwen3-235b-a22b-instruct-2507	alibaba	1465	±7	7.3K
98	grok-4.1	xai	1464	±9	5.3K
99	amazon-nova-experimental-chat-26-02-10	amazon	1464	±32	316
100	longcat-flash-chat-2602-exp	meituan	1464	±14	2.0K

能力评分数据来自 LMArena(原 LMSYS Chatbot Arena)公开发布的排行榜数据集「中文(chinese)」切片,许可 CC-BY-4.0,本站署名引用、未作修改。 Arena 分由真人匿名中文盲测对战经统计模型(Bradley-Terry)估算,反映相对偏好、非绝对能力,可能滞后或随样本波动,仅供参考。本栏目仅作信息整理,不构成任何投资或采购建议。本站不自建评测跑分。榜单发布于 2026-10-15。