模型排行

排名	变化	模型	得分	置信区间	投票	价格 $/M
1	—	kimi-k2-0711-preview开源Moonshot AI	1,421	±5	18,588	—
2	—	deepseek-v3.1开源DeepSeek · 另有 1 个档位	1,419	±9	4,844	—
3	—	qwen3-235b-a22b-instruct-2507开源Alibaba	1,419	±6	12,971	—
4	—	deepseek-r1-0528开源DeepSeek · 另有 1 个档位	1,417	±6	21,287	—
5	—	glm-4.5开源Z.ai	1,410	±6	11,120	—
6	—	qwen3-235b-a22b-thinking-2507开源Alibaba	1,399	±7	8,778	—
7	—	qwen3-235b-a22b-no-thinking开源Alibaba	1,399	±5	29,635	—
8	—	deepseek-v3-0324开源DeepSeek · 另有 1 个档位	1,392	±4	32,987	—
9	—	qwen3-coder-480b-a35b-instruct开源Alibaba	1,382	±6	11,708	—
10	—	qwen3-30b-a3b-instruct-2507开源Alibaba	1,379	±7	9,090	—
11	—	glm-4.5-air开源Z.ai	1,377	±6	10,099	—
12	—	qwen3-235b-a22b开源Alibaba	1,374	±5	25,684	—
13	—	minimax-m1开源MiniMax	1,368	±5	22,914	—
14	—	gemma-3-27b-it开源Google	1,364	±4	37,129	—
15	—	gpt-oss-120b开源OpenAI	1,355	±7	8,215	$0.26
16	—	mistral-small-2506开源Mistral	1,355	±6	14,824	—
17	—	step-3开源StepFun	1,354	±10	3,250	—
18	—	command-a-03-2025开源Cohere	1,347	±4	36,479	—
19	—	qwen3-32b开源Alibaba	1,346	±9	4,074	—
20	—	llama-3.1-nemotron-ultra-253b-v1开源NVIDIA	1,345	±11	2,656	—
21	—	glm-4.5v开源Z.ai	1,343	±16	1,297	—
22	—	nvidia-llama-3.3-nemotron-super-49b-v1.5开源NVIDIA	1,342	±9	3,898	—
23	—	gemma-3-12b-it开源Google	1,341	±9	3,976	—
24	—	qwq-32b开源Alibaba · 另有 1 个档位	1,337	±5	24,357	—
25	—	llama-3.1-405b-instruct-bf16开源Meta	1,335	±4	43,788	—
26	—	llama-3.1-405b-instruct-fp8开源Meta	1,333	±3	63,038	—
27	—	gpt-oss-20b开源OpenAI	1,330	±7	7,826	—
28	—	qwen3-30b-a3b开源Alibaba	1,329	±5	25,832	—
29	—	llama-4-maverick-17b-128e-instruct开源Meta	1,326	±5	30,323	—
30	—	llama-3.3-nemotron-49b-super-v1开源NVIDIA	1,325	±12	2,371	—
31	—	deepseek-v2.5-1210开源DeepSeek · 另有 1 个档位	1,322	±8	7,243	—
32	—	llama-4-scout-17b-16e-instruct开源Meta	1,320	±5	19,983	—
33	—	llama-3.3-70b-instruct开源Meta	1,320	±3	54,376	—
34	—	gemma-3n-e4b-it开源Google	1,319	±6	18,408	—
35	—	qwen-max-0919开源Alibaba	1,317	±5	17,432	—
36	—	athene-v2-chat开源NexusFlow	1,314	±4	26,074	—
37	—	mistral-large-2407开源Mistral · 另有 1 个档位	1,314	±4	48,217	—
38	—	athene-70b-0725开源NexusFlow	1,305	±5	20,580	—
39	—	gemma-3-4b-it开源Google	1,304	±9	4,321	—
40	—	qwen2.5-72b-instruct开源Alibaba	1,302	±4	41,519	—
41	—	mistral-small-3.1-24b-instruct-2503开源Mistral	1,300	±5	19,630	—
42	—	llama-3.1-nemotron-70b-instruct开源NVIDIA	1,298	±7	7,577	—
43	—	llama-3.1-70b-instruct开源Meta	1,295	±3	58,637	—
44	—	jamba-1.5-large开源AI21 Labs	1,289	±7	9,125	—
45	—	llama-3.1-tulu-3-70b开源Ai2	1,289	±10	3,010	—
46	—	gemma-2-27b-it开源Google	1,287	±3	79,538	—
47	—	llama-3.1-nemotron-51b-instruct开源NVIDIA	1,287	±10	3,889	—
48	—	gemma-2-9b-it-simpo开源Princeton	1,280	±7	10,548	—
49	—	nemotron-4-340b-instruct开源NVIDIA	1,279	±5	20,608	—
50	—	command-r-plus-08-2024开源Cohere	1,279	±6	10,535	—
51	—	llama-3-70b-instruct开源Meta	1,276	±3	163,629	—
52	—	mistral-small-24b-instruct-2501开源Mistral	1,276	±6	15,321	—
53	—	qwen2.5-coder-32b-instruct开源Alibaba	1,272	±8	5,730	—
54	—	c4ai-aya-expanse-32b开源Cohere	1,269	±5	28,768	—
55	—	gemma-2-9b-it开源Google	1,265	±4	57,197	—
56	—	command-r-plus开源Cohere	1,265	±4	80,846	—
57	—	deepseek-coder-v2开源DeepSeek	1,265	±6	15,753	—
58	—	qwen2-72b-instruct开源Alibaba	1,265	±5	38,872	—
59	—	phi-4开源Microsoft	1,258	±4	25,213	—
60	—	olmo-2-0325-32b-instruct开源Ai2	1,256	±11	3,460	—
61	—	command-r-08-2024开源Cohere	1,255	±6	10,851	—
62	—	jamba-1.5-mini开源AI21 Labs	1,241	±7	9,274	—
63	—	ministral-8b-2410开源Mistral	1,240	±9	5,111	—
64	—	qwen1.5-110b-chat开源Alibaba	1,237	±5	27,430	—
65	—	qwen1.5-72b-chat开源Alibaba	1,236	±5	40,658	—
66	—	mixtral-8x22b-instruct-v0.1开源Mistral	1,232	±4	53,751	—
67	—	command-r开源Cohere	1,231	±5	56,398	—
68	—	c4ai-aya-expanse-8b开源Cohere	1,228	±7	10,391	—
69	—	llama-3-8b-instruct开源Meta	1,225	±3	109,056	—
70	—	llama-3.1-tulu-3-8b开源Ai2	1,225	±10	3,074	—
71	—	yi-1.5-34b-chat开源01 AI	1,217	±5	25,135	—
72	—	zephyr-orpo-141b-A35b-v0.1开源HuggingFace	1,216	±10	4,854	—
73	—	llama-3.1-8b-instruct开源Meta	1,215	±4	52,578	—
74	—	granite-3.1-8b-instruct开源IBM	1,214	±10	3,289	—
75	—	qwen1.5-32b-chat开源Alibaba	1,209	±6	22,765	—
76	—	phi-3-medium-4k-instruct开源Microsoft	1,202	±5	26,105	—
77	—	gemma-2-2b-it开源Google	1,201	±4	48,892	—
78	—	mixtral-8x7b-instruct-v0.1开源Mistral	1,201	±4	76,126	—
79	—	internlm2_5-20b-chat开源InternLM	1,199	±7	10,599	—
80	—	dbrx-instruct-preview开源Databricks	1,199	±6	33,743	—
81	—	qwen1.5-14b-chat开源Alibaba	1,196	±7	18,687	—
82	—	granite-3.0-8b-instruct开源IBM	1,188	±8	7,002	—
83	—	wizardlm-70b开源Microsoft	1,188	±9	8,383	—
84	—	yi-34b-chat开源01 AI	1,187	±7	15,917	—
85	—	deepseek-llm-67b-chat开源DeepSeek	1,187	±11	4,988	—
86	—	openchat-3.5-0106开源OpenChat · 另有 1 个档位	1,186	±8	12,990	—
87	—	granite-3.1-2b-instruct开源IBM	1,185	±11	3,380	—
88	—	snowflake-arctic-instruct开源Snowflake	1,184	±6	34,173	—
89	—	tulu-2-dpo-70b开源AllenAI/UW	1,183	±10	6,658	—
90	—	gemma-1.1-7b-it开源Google	1,181	±6	25,070	—
91	—	openhermes-2.5-mistral-7b开源NousResearch	1,179	±10	5,088	—
92	—	vicuna-33b开源LMSYS	1,177	±6	22,936	—
93	—	starling-lm-7b-beta开源Nexusflow	1,177	±7	16,676	—
94	—	phi-3-small-8k-instruct开源Microsoft	1,176	±6	18,476	—
95	—	llama-2-70b-chat开源Meta	1,175	±5	39,595	—
96	—	llama-3.2-3b-instruct开源Meta	1,172	±7	8,390	—
97	—	starling-lm-7b-alpha开源UC Berkeley	1,172	±8	10,415	—
98	—	nous-hermes-2-mixtral-8x7b-dpo开源NousResearch	1,170	±12	3,836	—
99	—	granite-3.0-2b-instruct开源IBM	1,163	±8	7,191	—
100	—	llama2-70b-steerlm-chat开源NVIDIA	1,162	±13	3,636	—
101	—	solar-10.7b-instruct-v1.0开源Upstage AI	1,158	±13	4,286	—
102	—	dolphin-2.2.1-mistral-7b开源Cognitive Computations	1,157	±15	1,714	—
103	—	mistral-7b-instruct-v0.2开源Mistral	1,155	±6	20,067	—
104	—	mpt-30b-chat开源MosaicML	1,154	±12	2,644	—
105	—	wizardlm-13b开源Microsoft	1,153	±9	7,176	—
106	—	falcon-180b-chat开源TII	1,149	±17	1,327	—
107	—	phi-3-mini-4k-instruct-june-2024开源Microsoft	1,149	±6	12,808	—
108	—	qwen1.5-7b-chat开源Alibaba	1,148	±10	4,872	—
109	—	llama-2-13b-chat开源Meta	1,147	±7	19,722	—
110	—	vicuna-13b开源LMSYS	1,145	±7	19,775	—
111	—	qwen-14b-chat开源Alibaba	1,141	±11	5,065	—
112	—	codellama-34b-instruct开源Meta	1,141	±9	7,509	—
113	—	gemma-7b-it开源Google	1,137	±9	9,176	—
114	—	phi-3-mini-128k-instruct开源Microsoft	1,137	±7	21,622	—
115	—	zephyr-7b-beta开源HuggingFace	1,136	±9	11,321	—
116	—	phi-3-mini-4k-instruct开源Microsoft	1,135	±6	21,097	—
117	—	zephyr-7b-alpha开源HuggingFace	1,132	±16	1,811	—
118	—	guanaco-33b开源UW	1,131	±12	2,996	—
119	—	smollm2-1.7b-instruct开源HuggingFace	1,130	±13	2,375	—
120	—	codellama-70b-instruct开源Meta	1,125	±18	1,192	—
121	—	stripedhyena-nous-7b开源Together AI	1,124	±11	5,276	—
122	—	llama-3.2-1b-instruct开源Meta	1,122	±7	8,523	—
123	—	vicuna-7b开源LMSYS	1,118	±9	7,017	—
124	—	gemma-1.1-2b-it开源Google	1,117	±7	11,351	—
125	—	mistral-7b-instruct开源Mistral	1,114	±9	9,142	—
126	—	llama-2-7b-chat开源Meta	1,113	±7	14,532	—
127	—	qwen1.5-4b-chat开源Alibaba	1,095	±9	7,816	—
128	—	gemma-2b-it开源Google	1,094	±11	4,918	—
129	—	olmo-7b-instruct开源Ai2	1,080	±11	6,503	—
130	—	koala-13b开源UC Berkeley	1,073	±10	7,020	—
131	—	gpt4all-13b-snoozy开源Nomic AI	1,066	±15	1,788	—
132	—	alpaca-13b开源Stanford	1,065	±11	5,864	—
133	—	mpt-7b-chat开源MosaicML	1,064	±12	3,997	—
134	—	chatglm3-6b开源Tsinghua	1,060	±12	4,763	—
135	—	RWKV-4-Raven-14B开源RWKV	1,044	±11	4,920	—
136	—	chatglm2-6b开源Tsinghua	1,030	±13	2,713	—
137	—	oasst-pythia-12b开源OpenAssistant	1,023	±11	6,368	—
138	—	chatglm-6b开源Tsinghua	999	±13	4,983	—
139	—	fastchat-t5-3b开源LMSYS	993	±12	4,288	—
140	—	dolly-v2-12b开源Databricks	978	±14	3,480	—
141	—	llama-13b开源Meta	970	±16	2,446	—
142	—	stablelm-tuned-alpha-7b开源Stability AI	955	±13	3,336	—

其余 92 个模型

智力进步

每家厂商当时最强的模型在 Artificial Analysis 智力指数上的分数，按模型发布的月份画。

涨得最多 · Anthropic +57.2目前最高 · Anthropic 64.6最高与最低相差 18.1 分时间范围 2025年11月 – 2026年11月

64.6Anthropic60.0Google56.9OpenAI50.1Meta46.5Kimi