大模型能力榜

📊 全部資料 · 總覽

按 LMArena 競技場(Chatbot Arena)的真人盲測對戰評分,排出大模型的綜合能力高下(overall 全球總榜)。Arena 分越高,代表在真人匿名對比中越受偏好——它反映的是偏好、非絕對能力,且頭部幾名的分差常落在置信區間內(見下表「置信區間」列),更宜看作近似並列。資料歸因 LMArena,客觀呈現、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	1505	±6	15.4K
2	claude-opus-5-high	anthropic	1504	±5	31.6K
3	claude-opus-4-6-high	anthropic	1503	±3	72.1K
4	claude-opus-4-6	anthropic	1497	±3	76.1K
5	claude-fable-5	anthropic	1495	±5	25.8K
6	gemini-3.7-flash-high	google	1490	±8	5.7K
7	claude-opus-4-7-high	anthropic	1490	±4	60.1K
8	muse-spark-1.2 (xHigh)	meta	1488	±10	3.2K
9	gemini-3.5-flash-high	google	1483	±4	32.2K
10	claude-opus-4-7	anthropic	1483	±4	61.3K
11	gemini-3.1-pro-preview	google	1480	±3	101.2K
12	qwen3.8-max	alibaba	1480	±6	11.8K
13	gemini-3-pro	google	1480	±4	40.7K
14	muse-spark-1.1	meta	1478	±5	22.2K
15	gemini-3.6-flash-high	google	1477	±5	20.2K
16	kimi-k3-max	moonshot	1476	±6	16.6K
17	gemini-3.5-flash-medium	google	1475	±5	30.6K
18	glm-5.3-max	zai	1475	±8	5.8K
19	qwen3.7-max-preview	alibaba	1474	±10	3.7K
20	muse-spark	meta	1474	±6	13.6K
21	gpt-5.5-high	openai	1471	±4	61.6K
22	qwen3.5-max-preview	alibaba	1471	±5	21.5K
23	gpt-5.4-high	openai	1470	±4	60.6K
24	ernie-5.1	baidu	1468	±5	37.1K
25	glm-5.2-max	zai	1467	±5	32.5K
26	gemini-3-flash	google	1467	±4	30.2K
27	gpt-5.5	openai	1466	±4	63.0K
28	glm-5.3-flash	zai	1466	±12	2.4K
29	mimo-v2.5-pro	xiaomi	1465	±4	56.0K
30	glm-5.1	zai	1463	±4	44.0K
31	claude-opus-4-8-high	anthropic	1461	±4	46.8K
32	claude-sonnet-4-6	anthropic	1458	±4	66.4K
33	gemini-2.5-pro	google	1458	±2	122.6K
34	kimi-k2.6	moonshot	1455	±5	37.5K
35	qwen3.7-plus	alibaba	1455	±5	36.3K
36	gpt-5.6-sol-xhigh	openai	1454	±5	21.5K
37	gpt-5.4	openai	1453	±4	63.6K
38	claude-opus-4-8	anthropic	1452	±4	47.5K
39	grok-4.5	xai	1451	±5	24.2K
40	deepseek-v4-pro	deepseek	1451	±4	54.2K
41	grok-4.20-beta-0309-reasoning	xai	1451	±4	62.2K
42	claude-opus-4-5-20251101	anthropic	1450	±3	70.0K
43	grok-4.20-multi-agent-beta-0309	xai	1450	±4	60.8K
44	dola-seed-2.0-pro	bytedance	1448	±3	74.3K
45	amazon-nova-experimental-chat-26-02-10	amazon	1448	±10	3.4K
46	claude-opus-4-5-20251101-high-32k	anthropic	1447	±4	36.3K
47	gpt-5.6-terra-xhigh	openai	1447	±5	22.3K
48	qwen3.6-max-preview	alibaba	1446	±8	5.2K
49	glm-5	zai	1446	±4	27.6K
50	kimi-k2.5-thinking	moonshot	1446	±3	70.6K
51	deepseek-v4-pro-high-preview	deepseek	1445	±4	51.6K
52	nvidia-nemotron-3-ultra-550b-a55b-nvfp4	nvidia	1445	±7	10.7K
53	grok-4.20-beta1	xai	1445	±5	26.6K
54	ernie-5.0-0110	baidu	1444	±4	34.9K
55	grok-4.6-high	xai	1444	±10	3.5K
56	gemini-3-flash (thinking-minimal)	google	1442	±3	85.9K
57	ernie-5.0-preview-1203	baidu	1442	±7	9.6K
58	claude-sonnet-5-high	anthropic	1442	±5	29.5K
59	gpt-5.1-high	openai	1442	±4	40.3K
60	gemma-4-31b	google	1442	±8	5.9K
61	deepseek-v4-pro-high-20260813	deepseek	1441	±9	4.2K
62	hy3	tencent	1441	±8	6.5K
63	glm-4.6	zai	1440	±4	35.1K
64	qwen3-max-preview	alibaba	1439	±5	27.2K
65	gpt-5.2-chat-latest-20260210	openai	1439	±4	34.2K
66	inkling	thinky	1439	±5	20.4K
67	claude-sonnet-4-5-20250929	anthropic	1438	±3	79.7K
68	qwen3.5-397b-a17b	alibaba	1438	±3	71.9K
69	glm-5v-turbo	zai	1437	±7	9.4K
70	gemini-3.5-flash-lite	google	1437	±5	20.1K
71	qwen3.6-plus	alibaba	1437	±4	45.4K
72	grok-4.1-thinking	xai	1437	±3	64.2K
73	qwen3.8-27b	alibaba	1437	±9	5.0K
74	mimo-v2-pro	xiaomi	1437	±5	24.4K
75	grok-4.1	xai	1436	±3	66.4K
76	glm-4.7	zai	1436	±6	11.9K
77	gemma-4-26b-a4b	google	1435	±8	5.8K
78	minimax-m3	minimax	1434	±4	43.0K
79	claude-sonnet-4-5-20250929-high-32k	anthropic	1434	±3	81.0K
80	deepseek-v4-flash	deepseek	1432	±4	48.9K
81	ernie-5.0-preview-1022	baidu	1430	±9	4.7K
82	glm-4.5	zai	1430	±5	23.7K
83	gpt-5.6-luna-xhigh	openai	1429	±5	22.9K
84	chatgpt-4o-latest-20250326	openai	1429	±3	80.7K
85	mistral-large-3	mistral	1428	±3	64.0K
86	mimo-v2.5	xiaomi	1428	±4	44.5K
87	deepseek-r1-0528	deepseek	1427	±6	18.1K
88	longcat-flash-chat-2602-exp	meituan	1426	±5	28.0K
89	grok-3-preview-02-24	xai	1425	±4	32.4K
90	deepseek-v3.2	deepseek	1425	±4	46.5K
91	deepseek-v3.2-exp-thinking	deepseek	1425	±7	8.9K
92	mistral-medium-2508	mistral	1425	±3	92.8K
93	deepseek-v4-flash-high-preview	deepseek	1424	±4	48.7K
94	longcat-flash-chat	meituan	1423	±6	11.2K
95	mimo-v2-omni	xiaomi	1423	±6	19.4K
96	deepseek-v3.2-exp	deepseek	1423	±6	11.8K
97	gpt-5.1	openai	1423	±4	43.0K
98	mistral-medium-3.5	mistral	1421	±7	11.0K
99	amazon-nova-experimental-chat-12-10	amazon	1421	±10	3.6K
100	qwen3-vl-235b-a22b-instruct	alibaba	1420	±7	11.2K

中文能力評測參考

LMArena 總榜以英文對戰為主;想看中文場景,可切到上方「中文榜」(LMArena 中文競技場),或參考國內權威榜單(本站不轉載其資料,僅外鏈):

能力評分資料來自 LMArena(原 LMSYS Chatbot Arena)公開發布的排行榜資料集,許可 CC-BY-4.0,本站署名引用、未作修改。 Arena 分由真人匿名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。