性能衰退追踪器

检测AI模型可能出现的性能下降。此追踪器监控排名变动(排行榜上的位置变化),并标记在24小时或7天内显著下降的模型。"衰退分"越高,意味着越多的警告信号。

存在风险的模型

265

下降中 (7天)

111

不稳定

160

持续下降

按性能衰退风险评分排名的模型

存在风险的模型

265 个模型显示性能衰退迹象,按风险评分排名。更高的风险评分表示更令人担忧的性能趋势。

衰退分	模型	提供商	质量	24小时排名	7天排名	严重程度	状态
82	Hunyuan A13B InstructTencent	Tencent	72.1	-19	-29	high	fragile
77	o4 Mini HighOpenAI	OpenAI	85.0	-26	-23	high	fragile
67	Tongyi DeepResearch 30B A3BAlibaba	Alibaba	81.9	-22	-20	high	fragile
66	Trinity Miniarcee-ai	arcee-ai	82.4	-15	-23	high	fragile
66	Qwen3 Coder 30B A3B InstructAlibaba	Alibaba	72.2	-21	-20	high	fragile
63	MiniMax M2MiniMax	MiniMax	72.7	-6	-26	high	fragile
61	Qwen3 30B A3BAlibaba	Alibaba	71.3	-8	-24	high	fragile
56	MiniMax M2.1MiniMax	MiniMax	73.1	-9	-21	high	fragile
55	GPT-5.1-Codex-MaxOpenAI	OpenAI	85.0	-18	-16	high	fragile
55	Gemini 2.5 Pro Preview 05-06Google	Google	82.5	+7	-25	high	fragile
53	Gemini 2.5 Pro Preview 06-05Google	Google	84.1	0	-24	high	fragile
53	Aion-2.0aion-labs	aion-labs	69.2	-12	-18	high	fragile
52	MiniMax M1MiniMax	MiniMax	68.3	-7	-20	high	fragile
47	Gemini 2.5 Flash Lite Preview 09-2025Google	Google	83.6	+9	-21	high	fragile
45	Qwen3 14BAlibaba	Alibaba	71.2	-2	-19	high	fragile
44	Sonar Reasoning ProPerplexity	Perplexity	61.4	-5	-17	high	fragile
43	GPT-5.1 ChatOpenAI	OpenAI	85.0	-6	-16	high	fragile
43	Sonar Pro SearchPerplexity	Perplexity	85.0	-6	-16	high	fragile
43	Step 3.5 FlashStepFun	StepFun	73.2	+3	-19	high	fragile
43	GPT-4o-mini Search PreviewOpenAI	OpenAI	72.8	+9	-19	high	fragile
43	Solar Pro 3Upstage	Upstage	72.5	+1	-19	high	fragile
41	Llama 4 MaverickMeta	Meta	76.6	+14	-18	high	fragile
38	MiniMax M2.7MiniMax	MiniMax	83.0	-7	-13	high	fragile
38	Mercury 2Inception	Inception	81.3	-7	-13	high	fragile
38	GPT-5 ChatOpenAI	OpenAI	74.9	-3	-15	high	fragile
38	Qwen2.5 72B InstructAlibaba	Alibaba	52.3	-5	-14	high	fragile
37	Qwen3 235B A22B Instruct 2507Alibaba	Alibaba	69.8	+11	-16	high	fragile
37	Palmyra X5Writer	Writer	64.7	-10	-11	high	fragile
36	MiniMax M2.5MiniMax	MiniMax	76.0	-9	-11	high	fragile
35	Qwen3 Coder PlusAlibaba	Alibaba	78.4	-6	-12	high	fragile
35	Qwen3 Coder NextAlibaba	Alibaba	76.7	-10	-10	high	fragile
35	Llama 3.3 Nemotron Super 49B V1.5NVIDIA	NVIDIA	68.6	+6	-15	high	fragile
35	Qwen3 Next 80B A3B Instruct (free)Alibaba	Alibaba	66.9	+8	-15	high	fragile
35	Sonar ProPerplexity	Perplexity	62.9	-6	-12	high	fragile
35	Qwen VL PlusAlibaba	Alibaba	60.8	-4	-13	high	fragile
33	GPT-4o-miniOpenAI	OpenAI	64.4	+5	-14	high	fragile
33	Claude 3.5 HaikuAnthropic	Anthropic	62.4	-2	-13	high	fragile
32	R1DeepSeek	DeepSeek	68.2	-3	-12	high	fragile
31	Grok 4.20 Multi-Agent BetaxAI	xAI	82.2	+7	-13	high	fragile
31	DeepSeek V3.2 ExpDeepSeek	DeepSeek	77.2	-10	-8	high	fragile
31	Claude 3.5 SonnetAnthropic	Anthropic	65.7	-6	-10	high	fragile
31	Gemma 3n 2B (free)Google	Google	58.0	+5	-13	high	fragile
30	Claude Opus 4.1Anthropic	Anthropic	81.9	-25	+8	high	fragile
30	Trinity Mini (free)arcee-ai	arcee-ai	72.6	-22	-4	high	stable
30	Nova Pro 1.0Amazon	Amazon	58.0	-5	-10	high	fragile
30	LFM2-2.6BLiquid AI	Liquid AI	53.2	-13	-6	high	fragile
29	o4 MiniOpenAI	OpenAI	83.6	+14	-12	high	fragile
29	Qwen3 Coder FlashAlibaba	Alibaba	78.1	-8	-8	high	fragile
29	KAT-Coder-Pro V1Kuaishou	Kuaishou	77.4	+7	-12	high	fragile
29	LongCat Flash ChatMeituan	Meituan	72.6	-23	-3	high	stable
29	Gemma 3 27B (free)Google	Google	62.7	-12	-6	high	fragile
29	Gemma 3n 4B (free)Google	Google	55.3	+10	-12	high	fragile
29	LFM2-24B-A2BLiquid AI	Liquid AI	53.2	0	-12	high	fragile
27	Mistral Small 3.1 24BMistral AI	Mistral AI	66.0	+5	-11	high	fragile
26	GPT-5 CodexOpenAI	OpenAI	85.0	-21	+6	high	fragile
25	GPT-5.2 ChatOpenAI	OpenAI	82.9	-15	-5	high	stable
25	Ministral 3 14B 2512Mistral AI	Mistral AI	73.5	-20	+16	high	fragile
25	Nemotron 3 Nano 30B A3B (free)NVIDIA	NVIDIA	67.7	+2	-10	high	fragile
25	Gemma 3 27BGoogle	Google	63.4	+8	-10	high	fragile
25	GPT-4 TurboOpenAI	OpenAI	60.4	+4	-10	high	fragile
25	LFM2.5-1.2B-Thinking (free)Liquid AI	Liquid AI	59.0	+4	-10	high	fragile
24	Qwen3 VL 30B A3B ThinkingAlibaba	Alibaba	85.0	-19	+18	high	fragile
24	Qwen3 4B (free)Alibaba	Alibaba	62.8	-16	-4	high	stable
23	Kimi K2.5Moonshot AI	Moonshot AI	85.0	+11	-9	high	fragile
23	GPT-5.1-Codex-MiniOpenAI	OpenAI	85.0	-18	+13	high	fragile
23	Gemini 2.5 ProGoogle	Google	84.7	-18	+16	high	fragile
23	Qwen3 VL 30B A3B InstructAlibaba	Alibaba	80.9	0	-9	high	fragile
23	Gemini 2.5 FlashGoogle	Google	80.0	+17	-9	high	fragile
23	ERNIE 4.5 VL 28B A3BBaidu	Baidu	74.8	+5	-9	high	fragile
23	Ministral 3 8B 2512Mistral AI	Mistral AI	73.5	-18	+17	high	fragile
23	ERNIE 4.5 21B A3B ThinkingBaidu	Baidu	70.0	+9	-9	high	fragile
22	ERNIE 4.5 VL 424B A47B Baidu	Baidu	69.3	-14	-4	high	stable
22	ERNIE 4.5 300B A47B Baidu	Baidu	63.2	-17	+9	high	fragile
21	Claude Haiku 4.5Anthropic	Anthropic	83.0	-2	-7	high	fragile
21	Codestral 2508Mistral AI	Mistral AI	64.6	+12	-8	high	fragile
21	UI-TARS 7B ByteDance	ByteDance	62.5	+19	-8	high	fragile
20	Kimi K2 0711Moonshot AI	Moonshot AI	72.3	-3	-6	high	fragile
20	Nemotron Nano 9B V2 (free)NVIDIA	NVIDIA	71.4	-15	+17	high	fragile
19	GPT-5.4 NanoOpenAI	OpenAI	85.0	+15	-7	high	fragile
19	Grok 4 FastxAI	xAI	83.2	+19	-7	high	fragile
19	Qwen3.5 397B A17BAlibaba	Alibaba	81.8	-14	+8	high	fragile
19	Claude 3.7 SonnetAnthropic	Anthropic	77.0	-14	+6	high	fragile
19	Nemotron Nano 12B 2 VLNVIDIA	NVIDIA	72.6	-14	+13	high	fragile
19	R1 Distill Qwen 32BDeepSeek	DeepSeek	60.0	+5	-7	high	fragile
19	Mistral Small CreativeMistral AI	Mistral AI	59.0	-14	+7	high	fragile
19	Maestro Reasoningarcee-ai	arcee-ai	55.5	+6	-7	high	fragile
18	Qwen3.5-FlashAlibaba	Alibaba	79.4	-8	-5	high	stable
18	Mistral Large 3 2512Mistral AI	Mistral AI	73.5	-18	0	high	stable
18	Llama 4 ScoutMeta	Meta	71.9	-18	+4	high	stable
18	Mistral Medium 3Mistral AI	Mistral AI	64.8	-14	-2	high	stable
18	Nova Lite 1.0Amazon	Amazon	58.0	-12	-3	high	stable
17	Seed 1.6 FlashByteDance	ByteDance	85.0	-17	+2	high	stable
17	GPT-5 MiniOpenAI	OpenAI	79.0	+8	-6	high	fragile
17	Nova Premier 1.0Amazon	Amazon	77.8	-12	+10	high	fragile
17	DeepSeek V3.2 SpecialeDeepSeek	DeepSeek	77.1	+9	-6	high	fragile
17	o3 MiniOpenAI	OpenAI	73.3	-17	+1	high	stable
17	Qwen3 235B A22B Thinking 2507Alibaba	Alibaba	69.1	-9	-4	high	stable
17	Qwen-PlusAlibaba	Alibaba	64.9	+2	-6	high	fragile
17	Rnj 1 Instructessentialai	essentialai	64.8	+8	-6	high	fragile
17	Qwen3 Coder 480B A35BAlibaba	Alibaba	64.1	+7	-6	high	fragile
16	Qwen3 VL 8B InstructAlibaba	Alibaba	80.9	-11	+10	high	fragile
16	Nova 2 LiteAmazon	Amazon	72.7	-11	+12	high	fragile
16	Devstral MediumMistral AI	Mistral AI	62.5	-11	+6	high	fragile
15	Mistral Medium 3.1Mistral AI	Mistral AI	70.1	-10	+9	high	fragile
15	GPT AudioOpenAI	OpenAI	68.4	-10	+14	high	fragile
15	Devstral 2 2512Mistral AI	Mistral AI	67.7	-10	+14	high	fragile
15	Spotlightarcee-ai	arcee-ai	62.1	-15	+5	high	stable
15	Llama 3.1 70B InstructMeta	Meta	59.8	-10	+11	high	fragile
15	Llama Guard 4 12BMeta	Meta	58.9	-10	+9	high	fragile
14	Qwen Plus 0728 (thinking)Alibaba	Alibaba	82.7	-9	+7	high	fragile
14	Qwen3.5-122B-A10BAlibaba	Alibaba	79.7	-14	+2	high	stable
14	Mistral Small 4Mistral AI	Mistral AI	79.4	-9	+6	high	fragile
14	Kimi K2 0905Moonshot AI	Moonshot AI	65.5	-9	+18	high	fragile
14	Olmo 3.1 32B ThinkAllen AI	Allen AI	64.8	-9	+6	high	fragile
13	GPT-5 NanoOpenAI	OpenAI	75.5	-8	+18	high	fragile
13	Sonar Deep ResearchPerplexity	Perplexity	55.6	-7	-3	high	stable
13	Llama 3.1 Nemotron 70B InstructNVIDIA	NVIDIA	53.1	-13	+1	high	stable
12	o4 Mini Deep ResearchOpenAI	OpenAI	85.0	-7	+11	high	fragile
12	Qwen3 VL 235B A22B ThinkingAlibaba	Alibaba	77.3	-7	+9	high	fragile
12	Olmo 3 32B ThinkAllen AI	Allen AI	66.3	-6	-3	high	stable
12	Llama 3.2 11B Vision InstructMeta	Meta	54.2	-7	+7	high	fragile
11	Seed 1.6ByteDance	ByteDance	85.0	-6	+12	high	fragile
11	gpt-oss-120b (free)OpenAI	OpenAI	73.7	-6	+28	high	fragile
11	Claude 3.7 Sonnet (thinking)Anthropic	Anthropic	69.7	-6	+12	high	fragile
11	Qwen3 8BAlibaba	Alibaba	64.9	-6	+7	high	fragile
11	Mistral Small 3Mistral AI	Mistral AI	59.3	-11	+4	high	stable
11	Gemma 3 12B (free)Google	Google	55.0	-5	-3	high	stable
11	Command R7B (12-2024)Cohere	Cohere	44.6	-3	-4	high	stable
10	GPT-5.3-CodexOpenAI	OpenAI	85.0	-10	+3	high	stable
10	GPT-4.1OpenAI	OpenAI	77.3	-5	+10	high	fragile
10	Gemini 2.0 Flash LiteGoogle	Google	75.6	-5	+9	high	fragile
10	Ministral 3 3B 2512Mistral AI	Mistral AI	72.6	+1	-5	high	stable
10	gpt-oss-20bOpenAI	OpenAI	68.3	-10	+5	high	stable
10	Grok 3 Mini BetaxAI	xAI	66.0	0	-5	high	stable
10	Llama 3.3 70B InstructMeta	Meta	65.6	+18	-5	high	stable
10	R1 Distill Llama 70BDeepSeek	DeepSeek	60.8	+2	-5	high	stable
10	Qwen-Max Alibaba	Alibaba	58.6	0	-5	high	stable
10	LFM2.5-1.2B-Instruct (free)Liquid AI	Liquid AI	53.2	+4	-5	high	stable
10	QwQ 32BAlibaba	Alibaba	46.9	-2	-4	high	stable
10	Llama Guard 3 8BMeta	Meta	42.7	+8	-5	high	stable
10	GPT-3.5 Turbo (older v0613)OpenAI	OpenAI	38.0	0	-5	high	stable
10	Mixtral 8x22B InstructMistral AI	Mistral AI	37.0	-2	-4	high	stable
9	Claude Sonnet 4.6Anthropic	Anthropic	89.2	-1	-4	medium	stable
9	Qwen3 Next 80B A3B ThinkingAlibaba	Alibaba	72.6	-9	0	medium	stable
9	Grok 3 BetaxAI	xAI	63.3	-9	+5	medium	stable
9	Virtuoso Largearcee-ai	arcee-ai	62.1	-9	+5	medium	stable
9	Mistral LargeMistral AI	Mistral AI	54.1	-4	+6	medium	fragile
9	GPT-4o-mini (2024-07-18)OpenAI	OpenAI	53.6	-4	+8	medium	fragile
9	Mistral Large 2407Mistral AI	Mistral AI	52.9	-9	+2	medium	stable
9	Mixtral 8x7B InstructMistral AI	Mistral AI	42.4	-7	-1	medium	stable
9	GPT-3.5 Turbo 16kOpenAI	OpenAI	39.9	-4	+6	medium	fragile
8	GPT-5.2OpenAI	OpenAI	92.7	-2	-3	medium	stable
8	GPT-5 ProOpenAI	OpenAI	91.9	+2	-4	medium	stable
8	Claude Sonnet 4.5Anthropic	Anthropic	89.0	0	-4	medium	stable
8	Seed-2.0-LiteByteDance	ByteDance	85.0	-3	+7	medium	fragile
8	Gemini 3.1 Flash Lite PreviewGoogle	Google	81.9	+7	-4	medium	stable
8	Claude Opus 4Anthropic	Anthropic	81.6	-8	+2	medium	stable
8	Qwen3.5-35B-A3BAlibaba	Alibaba	78.3	-3	+7	medium	fragile
8	DeepSeek V3.2DeepSeek	DeepSeek	74.1	+19	-4	medium	stable
8	DeepSeek V3.1DeepSeek	DeepSeek	73.6	+17	-4	medium	stable
8	DeepSeek V3DeepSeek	DeepSeek	69.6	+17	-4	medium	stable
8	Gemma 2 27BGoogle	Google	59.5	-8	0	medium	stable
8	MiniMax M2-herMiniMax	MiniMax	59.4	-8	+1	medium	stable
8	Qwen2.5 VL 32B InstructAlibaba	Alibaba	56.5	+3	-4	medium	stable
8	Llama 3 8B InstructMeta	Meta	30.3	-2	-3	medium	stable
7	Grok 4.1 FastxAI	xAI	86.9	-1	-3	medium	stable
7	Seed-2.0-MiniByteDance	ByteDance	85.0	-2	+7	medium	fragile
7	Nemotron 3 Super (free)NVIDIA	NVIDIA	84.1	-2	+7	medium	fragile
7	MiMo-V2-FlashXiaomi	Xiaomi	82.6	-7	+5	medium	stable
7	Qwen2.5 Coder 32B InstructAlibaba	Alibaba	42.3	-7	0	medium	stable
7	Inflection 3 PiInflection	Inflection	36.6	-1	-3	medium	stable
6	GPT-5.1OpenAI	OpenAI	85.2	+36	-3	medium	stable
6	Qwen3 VL 32B InstructAlibaba	Alibaba	80.9	+14	-3	medium	stable
6	o1-proOpenAI	OpenAI	76.4	-6	+4	medium	stable
6	o3 Mini HighOpenAI	OpenAI	65.3	-6	+3	medium	stable
6	Phi 4Microsoft	Microsoft	59.5	+4	-3	medium	stable
6	Aion-1.0aion-labs	aion-labs	56.4	-1	+9	medium	fragile
6	GPT-4o (2024-05-13)OpenAI	OpenAI	52.6	+1	-3	medium	stable
6	Qwen2.5 Coder 7B InstructAlibaba	Alibaba	42.8	-1	+6	medium	fragile
5	Gemini 3 Flash PreviewGoogle	Google	89.4	+1	+7	medium	fragile
5	Grok 4xAI	xAI	85.7	+14	+8	medium	fragile
5	Gemini 3.1 Pro PreviewGoogle	Google	85.5	+19	+9	medium	fragile
5	o3OpenAI	OpenAI	85.5	+30	+20	medium	fragile
5	MiMo-V2-OmniXiaomi	Xiaomi	85.0	+22	+16	medium	fragile
5	MiMo-V2-ProXiaomi	Xiaomi	85.0	+2	+35	medium	fragile
5	GPT-5.3 ChatOpenAI	OpenAI	85.0	+31	+16	medium	fragile
5	GPT-5.2-CodexOpenAI	OpenAI	85.0	+3	+27	medium	fragile
5	GPT-5.1-CodexOpenAI	OpenAI	85.0	+16	+7	medium	fragile
5	Grok Code Fast 1xAI	xAI	84.6	+13	+7	medium	fragile
5	Nemotron Nano 12B 2 VL (free)NVIDIA	NVIDIA	82.3	+7	+6	medium	fragile
5	Qwen3 30B A3B Thinking 2507Alibaba	Alibaba	80.8	0	+9	medium	fragile
5	GPT-4.1 NanoOpenAI	OpenAI	80.5	+12	+12	medium	fragile
5	Qwen3.5-27BAlibaba	Alibaba	79.1	+17	+11	medium	fragile
5	Composer 2Cursor	Cursor	76.4	+10	+15	medium	fragile
5	Qwen3 MaxAlibaba	Alibaba	75.8	+19	+7	medium	fragile
5	Qwen3 30B A3B Instruct 2507Alibaba	Alibaba	75.0	+7	+22	medium	fragile
5	Gemini 2.0 FlashGoogle	Google	74.9	+3	+7	medium	fragile
5	gpt-oss-20b (free)OpenAI	OpenAI	73.7	+25	+20	medium	fragile
5	DeepSeek V3.1 TerminusDeepSeek	DeepSeek	73.6	+16	+15	medium	fragile
5	Nemotron 3 SuperNVIDIA	NVIDIA	73.5	+23	+13	medium	fragile
5	Trinity Large Preview (free)arcee-ai	arcee-ai	72.6	+16	+13	medium	fragile
5	Kimi K2 ThinkingMoonshot AI	Moonshot AI	72.6	+17	+15	medium	fragile
5	GPT-4o AudioOpenAI	OpenAI	71.9	+13	+8	medium	fragile
5	Nemotron Nano 9B V2NVIDIA	NVIDIA	71.4	+3	+6	medium	fragile
5	Qwen3 32BAlibaba	Alibaba	71.2	+8	+11	medium	fragile
5	Qwen3 235B A22BAlibaba	Alibaba	71.1	+8	+15	medium	fragile
5	Jamba Large 1.7AI21 Labs	AI21 Labs	71.0	+3	+6	medium	fragile
5	Qwen3 Next 80B A3B InstructAlibaba	Alibaba	70.0	+11	+10	medium	fragile
5	Qwen3 Coder 480B A35B (free)Alibaba	Alibaba	68.8	0	+10	medium	fragile
5	Mistral Small 3.2 24BMistral AI	Mistral AI	67.2	+15	+9	medium	fragile
5	Cogito v2.1 671Bdeepcogito	deepcogito	66.7	+10	+15	medium	fragile
5	ERNIE 4.5 21B A3BBaidu	Baidu	65.1	+1	+17	medium	fragile
5	Olmo 3.1 32B InstructAllen AI	Allen AI	64.9	+8	+22	medium	fragile
5	GPT-4o Search PreviewOpenAI	OpenAI	63.4	+7	+7	medium	fragile
5	MercuryInception	Inception	63.3	+12	+7	medium	fragile
5	Devstral Small 1.1Mistral AI	Mistral AI	62.5	+2	+9	medium	fragile
5	Mistral Small 3.1 24B (free)Mistral AI	Mistral AI	62.0	+2	+8	medium	fragile
5	Gemma 3 4B (free)Google	Google	60.9	+12	+7	medium	fragile
5	Qwen-TurboAlibaba	Alibaba	60.6	+12	+11	medium	fragile
5	Command ACohere	Cohere	59.9	0	+6	medium	fragile
5	Gemma 3 12BGoogle	Google	56.0	-5	+2	medium	stable
5	Granite 4.0 MicroIBM	IBM	55.1	+4	+6	medium	fragile
5	Gemma 3n 4BGoogle	Google	46.1	-5	0	medium	stable
4	Grok 4.20 BetaxAI	xAI	85.7	+13	-2	low	stable
4	Gemini 3.1 Pro Preview Custom ToolsGoogle	Google	85.0	+8	-2	low	stable
4	Gemini 2.5 Flash LiteGoogle	Google	81.3	-4	+4	low	stable
4	Reka Edgereka	reka	77.2	-4	+208	low	preliminary
4	Composer 2 FastCursor	Cursor	76.4	+8	-2	low	stable
4	Grok 3xAI	xAI	73.6	+12	-2	low	stable
4	Qwen VL MaxAlibaba	Alibaba	67.9	+1	-2	low	stable
4	gpt-oss-120bOpenAI	OpenAI	67.5	-4	+3	low	stable
4	GPT-4oOpenAI	OpenAI	64.2	+6	-2	low	stable
4	MiniMax-01MiniMax	MiniMax	61.8	+11	-2	low	stable
4	GPT-4o (2024-08-06)OpenAI	OpenAI	55.4	-4	+4	low	stable
4	Mistral Large 2411Mistral AI	Mistral AI	49.7	-4	0	low	stable
4	Llama 3.1 8B InstructMeta	Meta	42.2	+2	-2	low	stable
3	GPT-5.4OpenAI	OpenAI	94.0	-1	-1	low	stable
3	o3 Deep ResearchOpenAI	OpenAI	91.5	-1	-1	low	stable
3	Grok 3 MinixAI	xAI	76.1	-3	+5	low	stable
3	GPT Audio MiniOpenAI	OpenAI	68.4	-1	-1	low	stable
3	LFM2-8B-A1BLiquid AI	Liquid AI	53.2	-1	-1	low	stable
3	GPT-3.5 TurboOpenAI	OpenAI	39.9	-3	+2	low	stable
3	Llama 3.2 3B InstructMeta	Meta	35.8	-1	-1	low	stable
3	Llama 3.2 1B InstructMeta	Meta	31.8	-3	+1	low	stable
2	Qwen3.5 Plus 2026-02-15Alibaba	Alibaba	85.0	+2	-1	low	stable
2	Qwen3 Max ThinkingAlibaba	Alibaba	81.8	+14	-1	low	stable
2	Claude Sonnet 4Anthropic	Anthropic	79.7	-2	+2	low	stable
2	Step 3.5 Flash (free)StepFun	StepFun	78.2	+7	-1	low	stable
2	DeepSeek V3 0324DeepSeek	DeepSeek	73.1	+8	-1	low	stable
2	SonarPerplexity	Perplexity	53.5	0	-1	low	stable
2	Nova Micro 1.0Amazon	Amazon	51.0	0	-1	low	stable
2	SWE-1.5Windsurf	Windsurf	49.1	-2	+1	low	stable
2	Qwen2.5 7B InstructAlibaba	Alibaba	45.5	-2	+1	low	stable
2	Claude 3 HaikuAnthropic	Anthropic	43.0	+2	-1	low	stable
2	GPT-4 Turbo PreviewOpenAI	OpenAI	42.7	+5	-1	low	stable
2	GPT-4 Turbo (older v1106)OpenAI	OpenAI	42.7	0	-1	low	stable
2	Llama 3 70B InstructMeta	Meta	40.4	-2	0	low	stable
2	GPT-4 (older v0314)OpenAI	OpenAI	39.0	+4	-1	low	stable
2	autofixer-01Vercel	Vercel	38.8	-2	+3	low	stable
2	Llama 3.2 3B Instruct (free)Meta	Meta	35.0	0	-1	low	stable
1	GPT-5.4 MiniOpenAI	OpenAI	93.3	-1	+1	low	stable
1	Claude Opus 4.5Anthropic	Anthropic	90.4	-1	+4	low	stable
1	GPT-4.1 MiniOpenAI	OpenAI	77.3	-1	+2	low	stable
1	Olmo 2 32B InstructAllen AI	Allen AI	44.3	-1	0	low	stable
1	Gemma 2 9BGoogle	Google	30.1	-1	0	low	stable

稳定模型

35 个模型无下降且排名状态稳定。这些模型表现一致。

#	模型	提供商	评分	24h	7d	状态
1	GPT-5.4 ProOpenAI	OpenAI	94.0	+3	+4	stable
4	GPT-5.2 ProOpenAI	OpenAI	92.7	+1	+2	stable
6	Claude Opus 4.6Anthropic	Anthropic	92.1	0	+4	stable
10	GPT-5OpenAI	OpenAI	90.0	0	+1	stable
14	o3 ProOpenAI	OpenAI	87.5	+1	+1	stable
25	Qwen3.5-9BAlibaba	Alibaba	85.0	+1	+5	stable
40	Qwen3 VL 8B ThinkingAlibaba	Alibaba	85.0	+14	+1	stable
51	MiniMax M2.5 (free)MiniMax	MiniMax	83.4	+4	+4	stable
67	gpt-oss-safeguard-20bOpenAI	OpenAI	81.6	+12	+2	stable
88	R1 0528DeepSeek	DeepSeek	77.6	+15	+2	stable
97	Qwen Plus 0728Alibaba	Alibaba	76.8	0	+1	stable
108	o1OpenAI	OpenAI	75.5	+19	+3	stable
114	Qwen3 VL 235B A22B InstructAlibaba	Alibaba	73.9	+7	+5	stable
121	Nemotron 3 Nano 30B A3BNVIDIA	NVIDIA	73.5	+24	+1	stable
174	Mercury CoderInception	Inception	66.8	+16	+2	stable
200	GPT-4o (2024-11-20)OpenAI	OpenAI	63.1	+6	+5	stable
218	Qwen2.5 VL 72B InstructAlibaba	Alibaba	60.2	+13	+3	stable
233	Llama 3.1 Nemotron Ultra 253B v1NVIDIA	NVIDIA	57.4	+3	+2	stable
236	Aion-1.0-Miniaion-labs	aion-labs	56.4	+13	+2	stable
237	Gemma 3 4BGoogle	Google	56.0	+6	+4	stable

显示 35 个稳定模型中的前 20 个。

如何检测性能衰退

我们的性能衰退检测系统使用多种信号来识别可能正在下降的模型。

下降中 (7天)

7天排名变化超过-2位的模型。一周内持续下降超过两个名次,表明该模型可能正在被竞争对手超越或出现性能问题。

脆弱状态

被评分系统标记为"脆弱"的模型。这些模型的性能指标不一致或评分处于边界,评估数据的微小变化可能导致显著波动。

持续下降

在24小时和7天两个时间维度上均下降的模型。当模型在短期和中期窗口都在失去排名时,表明这是持续的下降趋势而非暂时波动。

风险评分

性能衰退风险评分综合了多个信号:7天排名下降权重2倍,24小时排名下降权重1倍,脆弱状态额外加5分。更高的评分表示有更大的性能衰退风险。

性能衰退追踪器

存在风险的模型

265

下降中 (7天)

111

不稳定

160

持续下降

按性能衰退风险评分排名的模型

存在风险的模型

265 个模型显示性能衰退迹象,按风险评分排名。更高的风险评分表示更令人担忧的性能趋势。

衰退分	模型	提供商	质量	24小时排名	7天排名	严重程度	状态
82	Hunyuan A13B InstructTencent	Tencent	72.1	-19	-29	high	fragile
77	o4 Mini HighOpenAI	OpenAI	85.0	-26	-23	high	fragile
67	Tongyi DeepResearch 30B A3BAlibaba	Alibaba	81.9	-22	-20	high	fragile
66	Trinity Miniarcee-ai	arcee-ai	82.4	-15	-23	high	fragile
66	Qwen3 Coder 30B A3B InstructAlibaba	Alibaba	72.2	-21	-20	high	fragile
63	MiniMax M2MiniMax	MiniMax	72.7	-6	-26	high	fragile
61	Qwen3 30B A3BAlibaba	Alibaba	71.3	-8	-24	high	fragile
56	MiniMax M2.1MiniMax	MiniMax	73.1	-9	-21	high	fragile
55	GPT-5.1-Codex-MaxOpenAI	OpenAI	85.0	-18	-16	high	fragile
55	Gemini 2.5 Pro Preview 05-06Google	Google	82.5	+7	-25	high	fragile
53	Gemini 2.5 Pro Preview 06-05Google	Google	84.1	0	-24	high	fragile
53	Aion-2.0aion-labs	aion-labs	69.2	-12	-18	high	fragile
52	MiniMax M1MiniMax	MiniMax	68.3	-7	-20	high	fragile
47	Gemini 2.5 Flash Lite Preview 09-2025Google	Google	83.6	+9	-21	high	fragile
45	Qwen3 14BAlibaba	Alibaba	71.2	-2	-19	high	fragile
44	Sonar Reasoning ProPerplexity	Perplexity	61.4	-5	-17	high	fragile
43	GPT-5.1 ChatOpenAI	OpenAI	85.0	-6	-16	high	fragile
43	Sonar Pro SearchPerplexity	Perplexity	85.0	-6	-16	high	fragile
43	Step 3.5 FlashStepFun	StepFun	73.2	+3	-19	high	fragile
43	GPT-4o-mini Search PreviewOpenAI	OpenAI	72.8	+9	-19	high	fragile
43	Solar Pro 3Upstage	Upstage	72.5	+1	-19	high	fragile
41	Llama 4 MaverickMeta	Meta	76.6	+14	-18	high	fragile
38	MiniMax M2.7MiniMax	MiniMax	83.0	-7	-13	high	fragile
38	Mercury 2Inception	Inception	81.3	-7	-13	high	fragile
38	GPT-5 ChatOpenAI	OpenAI	74.9	-3	-15	high	fragile
38	Qwen2.5 72B InstructAlibaba	Alibaba	52.3	-5	-14	high	fragile
37	Qwen3 235B A22B Instruct 2507Alibaba	Alibaba	69.8	+11	-16	high	fragile
37	Palmyra X5Writer	Writer	64.7	-10	-11	high	fragile
36	MiniMax M2.5MiniMax	MiniMax	76.0	-9	-11	high	fragile
35	Qwen3 Coder PlusAlibaba	Alibaba	78.4	-6	-12	high	fragile
35	Qwen3 Coder NextAlibaba	Alibaba	76.7	-10	-10	high	fragile
35	Llama 3.3 Nemotron Super 49B V1.5NVIDIA	NVIDIA	68.6	+6	-15	high	fragile
35	Qwen3 Next 80B A3B Instruct (free)Alibaba	Alibaba	66.9	+8	-15	high	fragile
35	Sonar ProPerplexity	Perplexity	62.9	-6	-12	high	fragile
35	Qwen VL PlusAlibaba	Alibaba	60.8	-4	-13	high	fragile
33	GPT-4o-miniOpenAI	OpenAI	64.4	+5	-14	high	fragile
33	Claude 3.5 HaikuAnthropic	Anthropic	62.4	-2	-13	high	fragile
32	R1DeepSeek	DeepSeek	68.2	-3	-12	high	fragile
31	Grok 4.20 Multi-Agent BetaxAI	xAI	82.2	+7	-13	high	fragile
31	DeepSeek V3.2 ExpDeepSeek	DeepSeek	77.2	-10	-8	high	fragile
31	Claude 3.5 SonnetAnthropic	Anthropic	65.7	-6	-10	high	fragile
31	Gemma 3n 2B (free)Google	Google	58.0	+5	-13	high	fragile
30	Claude Opus 4.1Anthropic	Anthropic	81.9	-25	+8	high	fragile
30	Trinity Mini (free)arcee-ai	arcee-ai	72.6	-22	-4	high	stable
30	Nova Pro 1.0Amazon	Amazon	58.0	-5	-10	high	fragile
30	LFM2-2.6BLiquid AI	Liquid AI	53.2	-13	-6	high	fragile
29	o4 MiniOpenAI	OpenAI	83.6	+14	-12	high	fragile
29	Qwen3 Coder FlashAlibaba	Alibaba	78.1	-8	-8	high	fragile
29	KAT-Coder-Pro V1Kuaishou	Kuaishou	77.4	+7	-12	high	fragile
29	LongCat Flash ChatMeituan	Meituan	72.6	-23	-3	high	stable
29	Gemma 3 27B (free)Google	Google	62.7	-12	-6	high	fragile
29	Gemma 3n 4B (free)Google	Google	55.3	+10	-12	high	fragile
29	LFM2-24B-A2BLiquid AI	Liquid AI	53.2	0	-12	high	fragile
27	Mistral Small 3.1 24BMistral AI	Mistral AI	66.0	+5	-11	high	fragile
26	GPT-5 CodexOpenAI	OpenAI	85.0	-21	+6	high	fragile
25	GPT-5.2 ChatOpenAI	OpenAI	82.9	-15	-5	high	stable
25	Ministral 3 14B 2512Mistral AI	Mistral AI	73.5	-20	+16	high	fragile
25	Nemotron 3 Nano 30B A3B (free)NVIDIA	NVIDIA	67.7	+2	-10	high	fragile
25	Gemma 3 27BGoogle	Google	63.4	+8	-10	high	fragile
25	GPT-4 TurboOpenAI	OpenAI	60.4	+4	-10	high	fragile
25	LFM2.5-1.2B-Thinking (free)Liquid AI	Liquid AI	59.0	+4	-10	high	fragile
24	Qwen3 VL 30B A3B ThinkingAlibaba	Alibaba	85.0	-19	+18	high	fragile
24	Qwen3 4B (free)Alibaba	Alibaba	62.8	-16	-4	high	stable
23	Kimi K2.5Moonshot AI	Moonshot AI	85.0	+11	-9	high	fragile
23	GPT-5.1-Codex-MiniOpenAI	OpenAI	85.0	-18	+13	high	fragile
23	Gemini 2.5 ProGoogle	Google	84.7	-18	+16	high	fragile
23	Qwen3 VL 30B A3B InstructAlibaba	Alibaba	80.9	0	-9	high	fragile
23	Gemini 2.5 FlashGoogle	Google	80.0	+17	-9	high	fragile
23	ERNIE 4.5 VL 28B A3BBaidu	Baidu	74.8	+5	-9	high	fragile
23	Ministral 3 8B 2512Mistral AI	Mistral AI	73.5	-18	+17	high	fragile
23	ERNIE 4.5 21B A3B ThinkingBaidu	Baidu	70.0	+9	-9	high	fragile
22	ERNIE 4.5 VL 424B A47B Baidu	Baidu	69.3	-14	-4	high	stable
22	ERNIE 4.5 300B A47B Baidu	Baidu	63.2	-17	+9	high	fragile
21	Claude Haiku 4.5Anthropic	Anthropic	83.0	-2	-7	high	fragile
21	Codestral 2508Mistral AI	Mistral AI	64.6	+12	-8	high	fragile
21	UI-TARS 7B ByteDance	ByteDance	62.5	+19	-8	high	fragile
20	Kimi K2 0711Moonshot AI	Moonshot AI	72.3	-3	-6	high	fragile
20	Nemotron Nano 9B V2 (free)NVIDIA	NVIDIA	71.4	-15	+17	high	fragile
19	GPT-5.4 NanoOpenAI	OpenAI	85.0	+15	-7	high	fragile
19	Grok 4 FastxAI	xAI	83.2	+19	-7	high	fragile
19	Qwen3.5 397B A17BAlibaba	Alibaba	81.8	-14	+8	high	fragile
19	Claude 3.7 SonnetAnthropic	Anthropic	77.0	-14	+6	high	fragile
19	Nemotron Nano 12B 2 VLNVIDIA	NVIDIA	72.6	-14	+13	high	fragile
19	R1 Distill Qwen 32BDeepSeek	DeepSeek	60.0	+5	-7	high	fragile
19	Mistral Small CreativeMistral AI	Mistral AI	59.0	-14	+7	high	fragile
19	Maestro Reasoningarcee-ai	arcee-ai	55.5	+6	-7	high	fragile
18	Qwen3.5-FlashAlibaba	Alibaba	79.4	-8	-5	high	stable
18	Mistral Large 3 2512Mistral AI	Mistral AI	73.5	-18	0	high	stable
18	Llama 4 ScoutMeta	Meta	71.9	-18	+4	high	stable
18	Mistral Medium 3Mistral AI	Mistral AI	64.8	-14	-2	high	stable
18	Nova Lite 1.0Amazon	Amazon	58.0	-12	-3	high	stable
17	Seed 1.6 FlashByteDance	ByteDance	85.0	-17	+2	high	stable
17	GPT-5 MiniOpenAI	OpenAI	79.0	+8	-6	high	fragile
17	Nova Premier 1.0Amazon	Amazon	77.8	-12	+10	high	fragile
17	DeepSeek V3.2 SpecialeDeepSeek	DeepSeek	77.1	+9	-6	high	fragile
17	o3 MiniOpenAI	OpenAI	73.3	-17	+1	high	stable
17	Qwen3 235B A22B Thinking 2507Alibaba	Alibaba	69.1	-9	-4	high	stable
17	Qwen-PlusAlibaba	Alibaba	64.9	+2	-6	high	fragile
17	Rnj 1 Instructessentialai	essentialai	64.8	+8	-6	high	fragile
17	Qwen3 Coder 480B A35BAlibaba	Alibaba	64.1	+7	-6	high	fragile
16	Qwen3 VL 8B InstructAlibaba	Alibaba	80.9	-11	+10	high	fragile
16	Nova 2 LiteAmazon	Amazon	72.7	-11	+12	high	fragile
16	Devstral MediumMistral AI	Mistral AI	62.5	-11	+6	high	fragile
15	Mistral Medium 3.1Mistral AI	Mistral AI	70.1	-10	+9	high	fragile
15	GPT AudioOpenAI	OpenAI	68.4	-10	+14	high	fragile
15	Devstral 2 2512Mistral AI	Mistral AI	67.7	-10	+14	high	fragile
15	Spotlightarcee-ai	arcee-ai	62.1	-15	+5	high	stable
15	Llama 3.1 70B InstructMeta	Meta	59.8	-10	+11	high	fragile
15	Llama Guard 4 12BMeta	Meta	58.9	-10	+9	high	fragile
14	Qwen Plus 0728 (thinking)Alibaba	Alibaba	82.7	-9	+7	high	fragile
14	Qwen3.5-122B-A10BAlibaba	Alibaba	79.7	-14	+2	high	stable
14	Mistral Small 4Mistral AI	Mistral AI	79.4	-9	+6	high	fragile
14	Kimi K2 0905Moonshot AI	Moonshot AI	65.5	-9	+18	high	fragile
14	Olmo 3.1 32B ThinkAllen AI	Allen AI	64.8	-9	+6	high	fragile
13	GPT-5 NanoOpenAI	OpenAI	75.5	-8	+18	high	fragile
13	Sonar Deep ResearchPerplexity	Perplexity	55.6	-7	-3	high	stable
13	Llama 3.1 Nemotron 70B InstructNVIDIA	NVIDIA	53.1	-13	+1	high	stable
12	o4 Mini Deep ResearchOpenAI	OpenAI	85.0	-7	+11	high	fragile
12	Qwen3 VL 235B A22B ThinkingAlibaba	Alibaba	77.3	-7	+9	high	fragile
12	Olmo 3 32B ThinkAllen AI	Allen AI	66.3	-6	-3	high	stable
12	Llama 3.2 11B Vision InstructMeta	Meta	54.2	-7	+7	high	fragile
11	Seed 1.6ByteDance	ByteDance	85.0	-6	+12	high	fragile
11	gpt-oss-120b (free)OpenAI	OpenAI	73.7	-6	+28	high	fragile
11	Claude 3.7 Sonnet (thinking)Anthropic	Anthropic	69.7	-6	+12	high	fragile
11	Qwen3 8BAlibaba	Alibaba	64.9	-6	+7	high	fragile
11	Mistral Small 3Mistral AI	Mistral AI	59.3	-11	+4	high	stable
11	Gemma 3 12B (free)Google	Google	55.0	-5	-3	high	stable
11	Command R7B (12-2024)Cohere	Cohere	44.6	-3	-4	high	stable
10	GPT-5.3-CodexOpenAI	OpenAI	85.0	-10	+3	high	stable
10	GPT-4.1OpenAI	OpenAI	77.3	-5	+10	high	fragile
10	Gemini 2.0 Flash LiteGoogle	Google	75.6	-5	+9	high	fragile
10	Ministral 3 3B 2512Mistral AI	Mistral AI	72.6	+1	-5	high	stable
10	gpt-oss-20bOpenAI	OpenAI	68.3	-10	+5	high	stable
10	Grok 3 Mini BetaxAI	xAI	66.0	0	-5	high	stable
10	Llama 3.3 70B InstructMeta	Meta	65.6	+18	-5	high	stable
10	R1 Distill Llama 70BDeepSeek	DeepSeek	60.8	+2	-5	high	stable
10	Qwen-Max Alibaba	Alibaba	58.6	0	-5	high	stable
10	LFM2.5-1.2B-Instruct (free)Liquid AI	Liquid AI	53.2	+4	-5	high	stable
10	QwQ 32BAlibaba	Alibaba	46.9	-2	-4	high	stable
10	Llama Guard 3 8BMeta	Meta	42.7	+8	-5	high	stable
10	GPT-3.5 Turbo (older v0613)OpenAI	OpenAI	38.0	0	-5	high	stable
10	Mixtral 8x22B InstructMistral AI	Mistral AI	37.0	-2	-4	high	stable
9	Claude Sonnet 4.6Anthropic	Anthropic	89.2	-1	-4	medium	stable
9	Qwen3 Next 80B A3B ThinkingAlibaba	Alibaba	72.6	-9	0	medium	stable
9	Grok 3 BetaxAI	xAI	63.3	-9	+5	medium	stable
9	Virtuoso Largearcee-ai	arcee-ai	62.1	-9	+5	medium	stable
9	Mistral LargeMistral AI	Mistral AI	54.1	-4	+6	medium	fragile
9	GPT-4o-mini (2024-07-18)OpenAI	OpenAI	53.6	-4	+8	medium	fragile
9	Mistral Large 2407Mistral AI	Mistral AI	52.9	-9	+2	medium	stable
9	Mixtral 8x7B InstructMistral AI	Mistral AI	42.4	-7	-1	medium	stable
9	GPT-3.5 Turbo 16kOpenAI	OpenAI	39.9	-4	+6	medium	fragile
8	GPT-5.2OpenAI	OpenAI	92.7	-2	-3	medium	stable
8	GPT-5 ProOpenAI	OpenAI	91.9	+2	-4	medium	stable
8	Claude Sonnet 4.5Anthropic	Anthropic	89.0	0	-4	medium	stable
8	Seed-2.0-LiteByteDance	ByteDance	85.0	-3	+7	medium	fragile
8	Gemini 3.1 Flash Lite PreviewGoogle	Google	81.9	+7	-4	medium	stable
8	Claude Opus 4Anthropic	Anthropic	81.6	-8	+2	medium	stable
8	Qwen3.5-35B-A3BAlibaba	Alibaba	78.3	-3	+7	medium	fragile
8	DeepSeek V3.2DeepSeek	DeepSeek	74.1	+19	-4	medium	stable
8	DeepSeek V3.1DeepSeek	DeepSeek	73.6	+17	-4	medium	stable
8	DeepSeek V3DeepSeek	DeepSeek	69.6	+17	-4	medium	stable
8	Gemma 2 27BGoogle	Google	59.5	-8	0	medium	stable
8	MiniMax M2-herMiniMax	MiniMax	59.4	-8	+1	medium	stable
8	Qwen2.5 VL 32B InstructAlibaba	Alibaba	56.5	+3	-4	medium	stable
8	Llama 3 8B InstructMeta	Meta	30.3	-2	-3	medium	stable
7	Grok 4.1 FastxAI	xAI	86.9	-1	-3	medium	stable
7	Seed-2.0-MiniByteDance	ByteDance	85.0	-2	+7	medium	fragile
7	Nemotron 3 Super (free)NVIDIA	NVIDIA	84.1	-2	+7	medium	fragile
7	MiMo-V2-FlashXiaomi	Xiaomi	82.6	-7	+5	medium	stable
7	Qwen2.5 Coder 32B InstructAlibaba	Alibaba	42.3	-7	0	medium	stable
7	Inflection 3 PiInflection	Inflection	36.6	-1	-3	medium	stable
6	GPT-5.1OpenAI	OpenAI	85.2	+36	-3	medium	stable
6	Qwen3 VL 32B InstructAlibaba	Alibaba	80.9	+14	-3	medium	stable
6	o1-proOpenAI	OpenAI	76.4	-6	+4	medium	stable
6	o3 Mini HighOpenAI	OpenAI	65.3	-6	+3	medium	stable
6	Phi 4Microsoft	Microsoft	59.5	+4	-3	medium	stable
6	Aion-1.0aion-labs	aion-labs	56.4	-1	+9	medium	fragile
6	GPT-4o (2024-05-13)OpenAI	OpenAI	52.6	+1	-3	medium	stable
6	Qwen2.5 Coder 7B InstructAlibaba	Alibaba	42.8	-1	+6	medium	fragile
5	Gemini 3 Flash PreviewGoogle	Google	89.4	+1	+7	medium	fragile
5	Grok 4xAI	xAI	85.7	+14	+8	medium	fragile
5	Gemini 3.1 Pro PreviewGoogle	Google	85.5	+19	+9	medium	fragile
5	o3OpenAI	OpenAI	85.5	+30	+20	medium	fragile
5	MiMo-V2-OmniXiaomi	Xiaomi	85.0	+22	+16	medium	fragile
5	MiMo-V2-ProXiaomi	Xiaomi	85.0	+2	+35	medium	fragile
5	GPT-5.3 ChatOpenAI	OpenAI	85.0	+31	+16	medium	fragile
5	GPT-5.2-CodexOpenAI	OpenAI	85.0	+3	+27	medium	fragile
5	GPT-5.1-CodexOpenAI	OpenAI	85.0	+16	+7	medium	fragile
5	Grok Code Fast 1xAI	xAI	84.6	+13	+7	medium	fragile
5	Nemotron Nano 12B 2 VL (free)NVIDIA	NVIDIA	82.3	+7	+6	medium	fragile
5	Qwen3 30B A3B Thinking 2507Alibaba	Alibaba	80.8	0	+9	medium	fragile
5	GPT-4.1 NanoOpenAI	OpenAI	80.5	+12	+12	medium	fragile
5	Qwen3.5-27BAlibaba	Alibaba	79.1	+17	+11	medium	fragile
5	Composer 2Cursor	Cursor	76.4	+10	+15	medium	fragile
5	Qwen3 MaxAlibaba	Alibaba	75.8	+19	+7	medium	fragile
5	Qwen3 30B A3B Instruct 2507Alibaba	Alibaba	75.0	+7	+22	medium	fragile
5	Gemini 2.0 FlashGoogle	Google	74.9	+3	+7	medium	fragile
5	gpt-oss-20b (free)OpenAI	OpenAI	73.7	+25	+20	medium	fragile
5	DeepSeek V3.1 TerminusDeepSeek	DeepSeek	73.6	+16	+15	medium	fragile
5	Nemotron 3 SuperNVIDIA	NVIDIA	73.5	+23	+13	medium	fragile
5	Trinity Large Preview (free)arcee-ai	arcee-ai	72.6	+16	+13	medium	fragile
5	Kimi K2 ThinkingMoonshot AI	Moonshot AI	72.6	+17	+15	medium	fragile
5	GPT-4o AudioOpenAI	OpenAI	71.9	+13	+8	medium	fragile
5	Nemotron Nano 9B V2NVIDIA	NVIDIA	71.4	+3	+6	medium	fragile
5	Qwen3 32BAlibaba	Alibaba	71.2	+8	+11	medium	fragile
5	Qwen3 235B A22BAlibaba	Alibaba	71.1	+8	+15	medium	fragile
5	Jamba Large 1.7AI21 Labs	AI21 Labs	71.0	+3	+6	medium	fragile
5	Qwen3 Next 80B A3B InstructAlibaba	Alibaba	70.0	+11	+10	medium	fragile
5	Qwen3 Coder 480B A35B (free)Alibaba	Alibaba	68.8	0	+10	medium	fragile
5	Mistral Small 3.2 24BMistral AI	Mistral AI	67.2	+15	+9	medium	fragile
5	Cogito v2.1 671Bdeepcogito	deepcogito	66.7	+10	+15	medium	fragile
5	ERNIE 4.5 21B A3BBaidu	Baidu	65.1	+1	+17	medium	fragile
5	Olmo 3.1 32B InstructAllen AI	Allen AI	64.9	+8	+22	medium	fragile
5	GPT-4o Search PreviewOpenAI	OpenAI	63.4	+7	+7	medium	fragile
5	MercuryInception	Inception	63.3	+12	+7	medium	fragile
5	Devstral Small 1.1Mistral AI	Mistral AI	62.5	+2	+9	medium	fragile
5	Mistral Small 3.1 24B (free)Mistral AI	Mistral AI	62.0	+2	+8	medium	fragile
5	Gemma 3 4B (free)Google	Google	60.9	+12	+7	medium	fragile
5	Qwen-TurboAlibaba	Alibaba	60.6	+12	+11	medium	fragile
5	Command ACohere	Cohere	59.9	0	+6	medium	fragile
5	Gemma 3 12BGoogle	Google	56.0	-5	+2	medium	stable
5	Granite 4.0 MicroIBM	IBM	55.1	+4	+6	medium	fragile
5	Gemma 3n 4BGoogle	Google	46.1	-5	0	medium	stable
4	Grok 4.20 BetaxAI	xAI	85.7	+13	-2	low	stable
4	Gemini 3.1 Pro Preview Custom ToolsGoogle	Google	85.0	+8	-2	low	stable
4	Gemini 2.5 Flash LiteGoogle	Google	81.3	-4	+4	low	stable
4	Reka Edgereka	reka	77.2	-4	+208	low	preliminary
4	Composer 2 FastCursor	Cursor	76.4	+8	-2	low	stable
4	Grok 3xAI	xAI	73.6	+12	-2	low	stable
4	Qwen VL MaxAlibaba	Alibaba	67.9	+1	-2	low	stable
4	gpt-oss-120bOpenAI	OpenAI	67.5	-4	+3	low	stable
4	GPT-4oOpenAI	OpenAI	64.2	+6	-2	low	stable
4	MiniMax-01MiniMax	MiniMax	61.8	+11	-2	low	stable
4	GPT-4o (2024-08-06)OpenAI	OpenAI	55.4	-4	+4	low	stable
4	Mistral Large 2411Mistral AI	Mistral AI	49.7	-4	0	low	stable
4	Llama 3.1 8B InstructMeta	Meta	42.2	+2	-2	low	stable
3	GPT-5.4OpenAI	OpenAI	94.0	-1	-1	low	stable
3	o3 Deep ResearchOpenAI	OpenAI	91.5	-1	-1	low	stable
3	Grok 3 MinixAI	xAI	76.1	-3	+5	low	stable
3	GPT Audio MiniOpenAI	OpenAI	68.4	-1	-1	low	stable
3	LFM2-8B-A1BLiquid AI	Liquid AI	53.2	-1	-1	low	stable
3	GPT-3.5 TurboOpenAI	OpenAI	39.9	-3	+2	low	stable
3	Llama 3.2 3B InstructMeta	Meta	35.8	-1	-1	low	stable
3	Llama 3.2 1B InstructMeta	Meta	31.8	-3	+1	low	stable
2	Qwen3.5 Plus 2026-02-15Alibaba	Alibaba	85.0	+2	-1	low	stable
2	Qwen3 Max ThinkingAlibaba	Alibaba	81.8	+14	-1	low	stable
2	Claude Sonnet 4Anthropic	Anthropic	79.7	-2	+2	low	stable
2	Step 3.5 Flash (free)StepFun	StepFun	78.2	+7	-1	low	stable
2	DeepSeek V3 0324DeepSeek	DeepSeek	73.1	+8	-1	low	stable
2	SonarPerplexity	Perplexity	53.5	0	-1	low	stable
2	Nova Micro 1.0Amazon	Amazon	51.0	0	-1	low	stable
2	SWE-1.5Windsurf	Windsurf	49.1	-2	+1	low	stable
2	Qwen2.5 7B InstructAlibaba	Alibaba	45.5	-2	+1	low	stable
2	Claude 3 HaikuAnthropic	Anthropic	43.0	+2	-1	low	stable
2	GPT-4 Turbo PreviewOpenAI	OpenAI	42.7	+5	-1	low	stable
2	GPT-4 Turbo (older v1106)OpenAI	OpenAI	42.7	0	-1	low	stable
2	Llama 3 70B InstructMeta	Meta	40.4	-2	0	low	stable
2	GPT-4 (older v0314)OpenAI	OpenAI	39.0	+4	-1	low	stable
2	autofixer-01Vercel	Vercel	38.8	-2	+3	low	stable
2	Llama 3.2 3B Instruct (free)Meta	Meta	35.0	0	-1	low	stable
1	GPT-5.4 MiniOpenAI	OpenAI	93.3	-1	+1	low	stable
1	Claude Opus 4.5Anthropic	Anthropic	90.4	-1	+4	low	stable
1	GPT-4.1 MiniOpenAI	OpenAI	77.3	-1	+2	low	stable
1	Olmo 2 32B InstructAllen AI	Allen AI	44.3	-1	0	low	stable
1	Gemma 2 9BGoogle	Google	30.1	-1	0	low	stable

稳定模型

35 个模型无下降且排名状态稳定。这些模型表现一致。

#	模型	提供商	评分	24h	7d	状态
1	GPT-5.4 ProOpenAI	OpenAI	94.0	+3	+4	stable
4	GPT-5.2 ProOpenAI	OpenAI	92.7	+1	+2	stable
6	Claude Opus 4.6Anthropic	Anthropic	92.1	0	+4	stable
10	GPT-5OpenAI	OpenAI	90.0	0	+1	stable
14	o3 ProOpenAI	OpenAI	87.5	+1	+1	stable
25	Qwen3.5-9BAlibaba	Alibaba	85.0	+1	+5	stable
40	Qwen3 VL 8B ThinkingAlibaba	Alibaba	85.0	+14	+1	stable
51	MiniMax M2.5 (free)MiniMax	MiniMax	83.4	+4	+4	stable
67	gpt-oss-safeguard-20bOpenAI	OpenAI	81.6	+12	+2	stable
88	R1 0528DeepSeek	DeepSeek	77.6	+15	+2	stable
97	Qwen Plus 0728Alibaba	Alibaba	76.8	0	+1	stable
108	o1OpenAI	OpenAI	75.5	+19	+3	stable
114	Qwen3 VL 235B A22B InstructAlibaba	Alibaba	73.9	+7	+5	stable
121	Nemotron 3 Nano 30B A3BNVIDIA	NVIDIA	73.5	+24	+1	stable
174	Mercury CoderInception	Inception	66.8	+16	+2	stable
200	GPT-4o (2024-11-20)OpenAI	OpenAI	63.1	+6	+5	stable
218	Qwen2.5 VL 72B InstructAlibaba	Alibaba	60.2	+13	+3	stable
233	Llama 3.1 Nemotron Ultra 253B v1NVIDIA	NVIDIA	57.4	+3	+2	stable
236	Aion-1.0-Miniaion-labs	aion-labs	56.4	+13	+2	stable
237	Gemma 3 4BGoogle	Google	56.0	+6	+4	stable

显示 35 个稳定模型中的前 20 个。

如何检测性能衰退

我们的性能衰退检测系统使用多种信号来识别可能正在下降的模型。

下降中 (7天)

7天排名变化超过-2位的模型。一周内持续下降超过两个名次,表明该模型可能正在被竞争对手超越或出现性能问题。

脆弱状态

被评分系统标记为"脆弱"的模型。这些模型的性能指标不一致或评分处于边界,评估数据的微小变化可能导致显著波动。

持续下降

在24小时和7天两个时间维度上均下降的模型。当模型在短期和中期窗口都在失去排名时,表明这是持续的下降趋势而非暂时波动。

风险评分

性能衰退风险评分综合了多个信号:7天排名下降权重2倍,24小时排名下降权重1倍,脆弱状态额外加5分。更高的评分表示有更大的性能衰退风险。

性能衰退追踪器

按性能衰退风险评分排名的模型

存在风险的模型

稳定模型

如何检测性能衰退

下降中 (7天)

脆弱状态

持续下降

风险评分

相关

性能衰退追踪器

按性能衰退风险评分排名的模型

存在风险的模型

稳定模型

如何检测性能衰退

下降中 (7天)

脆弱状态

持续下降

风险评分

相关