综合能力排行榜

融合多平台综合评测，汇成一个 AISOTA 综合分。

数据截止 2026-08-1448 个模型

排名	模型	Artificial Analysis	LiveBench	AISOTA 分
1SOTA	Claude Fable 5 (with fallback)2 sources	90.0	97.7	93.8
2	gpt-5.5-xhigh	—	93.0	93.0
3	Claude Opus 5 (max)2 sources	95.0	90.7	92.8
4	GPT-5.6 Sol (max)2 sources	85.0	95.3	90.2
5	smaug-agentic	—	86.0	86.0
6	Grok 4.6 (high)2 sources	80.0	79.1	79.5
7	Kimi K3 (max)2 sources	75.0	81.4	78.2
8	Qwen3.8 Max2 sources	70.0	83.7	76.8
9	gpt-5.4-xhigh	—	74.4	74.4
10	Gemini 3.7 Flash (high)2 sources	55.0	88.4	71.7
11	Muse Spark 1.2 (xhigh)2 sources	65.0	76.7	70.8
12	gemini-3.1-pro-preview-high	—	67.4	67.4
13	GPT-5.6 Terra (max)2 sources	60.0	72.1	66.0
14	claude-opus-4-8-max-effort	—	65.1	65.1
15	grok-4.5	—	62.8	62.8
16	claude-opus-4-7-xhigh-effort	—	60.5	60.5
17	DeepSeek V4 Pro 0813 (max)2 sources	50.0	69.8	59.9
18	claude-sonnet-5-xhigh-effort	—	58.1	58.1
19	muse-spark-1.1-xhigh	—	55.8	55.8
20	gemini-3.5-flash-high	—	53.5	53.5
21	gpt-5.2-2025-12-11-high	—	51.2	51.2
22	claude-opus-4-6-thinking-auto-high-effort	—	48.8	48.8
23	deepseek-v4-flash-0731	—	46.5	46.5
24	gemini-3.6-flash-high	—	44.2	44.2
25	qwen3.7-max	—	41.9	41.9
26	gpt-5.2-codex	—	39.5	39.5
27	GLM-5.2 (max)2 sources	45.0	32.6	38.8
28	GPT-5.6 Luna (max)2 sources	40.0	37.2	38.6
29	Motif 3	35.0	—	35.0
30	claude-sonnet-4-6-thinking-auto-medium-effort	—	34.9	34.9
31	claude-opus-4-5-20251101-thinking-64k-high-effort	—	30.2	30.2
32	Inkling2 sources	25.0	27.9	26.4
33	deepseek-v4-pro	—	25.6	25.6
34	kimi-k2.6-thinking	—	23.3	23.3
35	gpt-5.4-nano-xhigh	—	20.9	20.9
36	MiniMax-M32 sources	30.0	11.6	20.8
37	Nemotron 3 Ultra	20.0	—	20.0
38	qwen3.6-plus	—	18.6	18.6
39	kimi-k2.7-code	—	16.3	16.3
40	grok-build-0.1	—	14.0	14.0
41	Solar Open2 250B	10.0	—	10.0
42	gpt-5.4-mini-xhigh	—	9.3	9.3
43	Gemini 3.5 Flash-Lite2 sources	15.0	2.3	8.7
44	deepseek-v4-flash	—	7.0	7.0
45	Muse Glimmer (high)	5.0	—	5.0
46	qwen3.6-27b	—	4.7	4.7
47	A.X-K2	0.0	—	0.0
48	grok-4.3	—	0.0	0.0