中文大模型能力榜

📊 全部資料 · 總覽

按 LMArena 中文競技場(Chinese Arena)的真人中文盲測對戰評分,排出大模型在中文場景的能力高下。它只統計中文對話裡的偏好勝負,和上方「總榜」(以英文為主)可能明顯不同——國產模型在這裡通常更靠前。評分反映偏好、非絕對能力,頭部分差常落在置信區間內(見下表),宜看作近似並列。資料歸因 LMArena,不構成投資或採購建議。榜單釋出於 2026-08-27

排名	模型	機構	Arena 分	置信區間	投票數
1	claude-opus-5-max	anthropic	1590	±19	1.1K
2	claude-opus-5-high	anthropic	1580	±14	2.2K
3	gemini-3.7-flash-high	google	1561	±31	389
4	claude-opus-4-6-high	anthropic	1551	±10	4.5K
5	claude-opus-4-6	anthropic	1547	±10	4.5K
6	claude-fable-5	anthropic	1544	±16	1.6K
7	qwen3.8-max	alibaba	1539	±22	732
8	gpt-5.5	openai	1536	±11	3.4K
9	claude-opus-4-7-high	anthropic	1534	±11	3.4K
10	gemini-3.6-flash-high	google	1533	±17	1.3K
11	gemini-3.5-flash-high	google	1533	±14	2.0K
12	qwen3.5-max-preview	alibaba	1532	±16	1.4K
13	gemini-3.1-pro-preview	google	1531	±9	6.1K
14	kimi-k2.6	moonshot	1529	±14	1.9K
15	grok-4.6-high	xai	1528	±34	293
16	qwen3.7-max-preview	alibaba	1526	±37	263
17	claude-opus-4-7	anthropic	1526	±11	3.5K
18	gemini-3-pro	google	1525	±11	3.2K
19	gpt-5.5-high	openai	1524	±11	3.5K
20	gpt-5.6-sol-xhigh	openai	1524	±16	1.4K
21	muse-spark-1.1	meta	1523	±16	1.5K
22	glm-5.3-max	zai	1520	±31	369
23	glm-5.1	zai	1518	±12	2.7K
24	gemini-3.5-flash-medium	google	1518	±14	1.9K
25	gpt-5.4-high	openai	1517	±11	3.5K
26	muse-spark-1.2 (xHigh)	meta	1516	±38	260
27	glm-5.2-max	zai	1516	±14	2.0K
28	kimi-k3-max	moonshot	1515	±18	1.1K
29	gpt-5.6-terra-xhigh	openai	1514	±16	1.5K
30	qwen3.8-27b	alibaba	1513	±34	291
31	glm-5	zai	1510	±14	1.7K
32	ernie-5.0-preview-1022	baidu	1509	±30	381
33	gemini-2.5-pro	google	1509	±7	8.8K
34	gemini-3-flash	google	1509	±13	2.3K
35	gpt-5.4	openai	1508	±11	3.7K
36	qwen3.7-plus	alibaba	1508	±14	2.0K
37	mimo-v2.5-pro	xiaomi	1503	±11	3.0K
38	muse-spark	meta	1503	±21	850
39	claude-opus-4-8	anthropic	1503	±12	2.7K
40	hy3	tencent	1502	±28	434
41	ernie-5.1	baidu	1502	±15	1.8K
42	grok-4.5	xai	1502	±16	1.6K
43	glm-4.6	zai	1500	±12	2.6K
44	ernie-5.0-0110	baidu	1499	±13	2.2K
45	claude-opus-4-8-high	anthropic	1498	±12	2.7K
46	kimi-k2.5-thinking	moonshot	1498	±10	3.9K
47	qwen3.5-397b-a17b	alibaba	1497	±10	4.3K
48	ernie-5.0-preview-1203	baidu	1497	±21	792
49	glm-4.7	zai	1495	±19	961
50	inkling	thinky	1495	±16	1.5K
51	gpt-5.1-high	openai	1493	±11	3.1K
52	nvidia-nemotron-3-ultra-550b-a55b-nvfp4	nvidia	1491	±23	631
53	glm-5v-turbo	zai	1491	±24	605
54	qwen3.6-max-preview	alibaba	1489	±33	320
55	dola-seed-2.0-pro	bytedance	1489	±10	4.4K
56	claude-sonnet-4-6	anthropic	1489	±10	3.8K
57	deepseek-v4-pro	deepseek	1485	±12	2.9K
58	claude-sonnet-5-high	anthropic	1483	±14	2.0K
59	deepseek-v4-pro-high-preview	deepseek	1483	±12	2.8K
60	gemma-4-26b-a4b	google	1483	±26	469
61	qwen3-235b-a22b-thinking-2507	alibaba	1483	±26	540
62	grok-4.20-beta-0309-reasoning	xai	1481	±11	3.6K
63	deepseek-v4-pro-high-20260813	deepseek	1481	±35	268
64	qwen3-max-preview	alibaba	1479	±13	2.1K
65	grok-4.20-multi-agent-beta-0309	xai	1478	±11	3.4K
66	grok-4.20-beta1	xai	1477	±16	1.5K
67	gemini-3-flash (thinking-minimal)	google	1477	±9	5.1K
68	qwen3.6-plus	alibaba	1473	±13	2.2K
69	qwen3.5-27b	alibaba	1472	±14	1.8K
70	gemma-4-31b	google	1472	±26	465
71	mimo-v2.5	xiaomi	1470	±13	2.1K
72	deepseek-v3.1-thinking	deepseek	1470	±18	1.2K
73	gpt-5.1	openai	1469	±10	3.3K
74	claude-opus-4-5-20251101	anthropic	1469	±9	4.8K
75	gpt-5.6-luna-xhigh	openai	1468	±16	1.6K
76	gemini-3.5-flash-lite	google	1468	±17	1.4K
77	mimo-v2-omni	xiaomi	1467	±20	859
78	deepseek-v4-flash-high-preview	deepseek	1467	±12	2.5K
79	glm-4.5	zai	1466	±14	1.8K
80	amazon-nova-experimental-chat-11-10	amazon	1466	±14	1.8K
81	qwen3.5-122b-a10b	alibaba	1466	±14	1.9K
82	longcat-flash-chat-2602-exp	meituan	1465	±15	1.7K
83	grok-4.1	xai	1465	±9	4.8K
84	mimo-v2-flash (non-thinking)	xiaomi	1465	±11	3.0K
85	deepseek-v4-flash	deepseek	1465	±12	2.5K
86	claude-opus-4-5-20251101-high-32k	anthropic	1465	±11	2.8K
87	qwen3-235b-a22b-instruct-2507	alibaba	1465	±8	6.8K
88	deepseek-v3.2-exp	deepseek	1463	±18	997
89	gemini-3.1-flash-lite-preview	google	1463	±10	3.6K
90	qwen3-vl-235b-a22b-instruct	alibaba	1462	±24	701
91	qwen3-next-80b-a3b-instruct	alibaba	1462	±14	1.7K
92	gpt-5.2-chat-latest-20260210	openai	1461	±13	2.1K
93	qwen3.5-35b-a3b	alibaba	1461	±14	1.8K
94	kimi-k2.5-instant	moonshot	1461	±27	454
95	deepseek-v3.2-exp-thinking	deepseek	1460	±26	500
96	claude-sonnet-4-5-20250929	anthropic	1459	±9	5.3K
97	claude-sonnet-4-5-20250929-high-32k	anthropic	1458	±9	5.3K
98	grok-4.1-thinking	xai	1458	±9	4.6K
99	grok-4-fast-chat	xai	1457	±28	463
100	Inkling Small	thinky	1457	±19	955

能力評分資料來自 LMArena(原 LMSYS Chatbot Arena)公開發布的排行榜資料集「中文(chinese)」切片,許可 CC-BY-4.0,本站署名引用、未作修改。 Arena 分由真人匿名中文盲測對戰經統計模型(Bradley-Terry)估算,反映相對偏好、非絕對能力,可能滯後或隨樣本波動,僅供參考。本欄目僅作資訊整理,不構成任何投資或採購建議。本站不自建評測跑分。榜單釋出於 2026-08-27。