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

285

下降中 (7天)

259

不稳定

252

持续下降

115

按性能衰退风险评分排名的模型

LMMarketCap.com

存在风险的模型

285 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
47	Qwen3.7 FlashAlibaba	Alibaba	40.0	-6	-18	high	fragile
46	Inkling Smallthinkingmachines	thinkingmachines	40.0	-5	-18	high	fragile
45	Llama 3.1 70B InstructMeta	Meta	63.6	0	-20	high	fragile
45	Claude Opus 5 (batch)Anthropic	Anthropic	40.0	-6	-17	high	fragile
45	Ling 3.0 Flashinclusionai	inclusionai	40.0	-6	-17	high	fragile
45	Gemini 3.6 FlashGoogle	Google	40.0	-6	-17	high	fragile
45	Gemini 3.6 Flash (batch)Google	Google	40.0	-6	-17	high	fragile
45	LongCat 2.0Meituan	Meituan	40.0	-6	-17	high	fragile
45	Inklingthinkingmachines	thinkingmachines	40.0	-6	-17	high	fragile
43	Muse Spark 1.2meta	meta	40.0	-4	-17	high	fragile
43	DeepSeek V4 Flash Latest~deepseek	~deepseek	40.0	-4	-17	high	fragile
43	DeepSeek V4 Flash 0731DeepSeek	DeepSeek	40.0	-4	-17	high	fragile
43	Kimi K3Moonshot AI	Moonshot AI	40.0	-6	-16	high	fragile
43	Kimi K3 (batch)Moonshot AI	Moonshot AI	40.0	-6	-16	high	fragile
43	Muse Spark 1.1meta	meta	40.0	-6	-16	high	fragile
43	Grok Latest~x-ai	~x-ai	40.0	-6	-16	high	fragile
43	Aion-3.0-Miniaion-labs	aion-labs	40.0	-6	-16	high	fragile
43	Aion-3.0aion-labs	aion-labs	40.0	-6	-16	high	fragile
43	Hy3Tencent	Tencent	40.0	-6	-16	high	fragile
39	MiniMax M2.7MiniMax	MiniMax	63.6	0	-17	high	fragile
38	Dots3-Note Preview (free)dots-studio	dots-studio	40.0	-3	-15	high	fragile
38	Gemini 3.7 FlashGoogle	Google	40.0	-3	-15	high	fragile
38	Gemini 3.7 Flash (batch)Google	Google	40.0	-3	-15	high	fragile
38	Seed 2.1 TurboByteDance	ByteDance	40.0	-3	-15	high	fragile
38	Qwen3.8 2.4T A95BAlibaba	Alibaba	40.0	-3	-15	high	fragile
38	Seed-2.0-CodeByteDance	ByteDance	40.0	-3	-15	high	fragile
38	DeepSeek V4 Pro 0813DeepSeek	DeepSeek	40.0	-3	-15	high	fragile
38	LFM2.5-2.6BLiquid AI	Liquid AI	40.0	-3	-15	high	fragile
38	LFM2.5-2.6B (free)Liquid AI	Liquid AI	40.0	-3	-15	high	fragile
38	Nemotron 3.5 LightningNVIDIA	NVIDIA	40.0	-3	-15	high	fragile
38	Nemotron 3.5 Lightning (free)NVIDIA	NVIDIA	40.0	-3	-15	high	fragile
38	Sakana Namazusakana	sakana	40.0	-3	-15	high	fragile
38	Solar Pro 4Upstage	Upstage	40.0	-3	-15	high	fragile
38	Muse Glimmer 30Bmeta	meta	40.0	-3	-15	high	fragile
35	Particle-1particle-ai	particle-ai	40.0	-2	-14	high	fragile
35	GPT-6.1 Terra ProOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6.1 Terra Pro (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6.1 TerraOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6.1 Terra (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	Kimi K3.1 ProMoonshot AI	Moonshot AI	40.0	-2	-14	high	fragile
35	Flint (free)recursive	recursive	40.0	-2	-14	high	fragile
35	MiniMax M3.1 FlashMiniMax	MiniMax	40.0	-2	-14	high	fragile
35	DeepSeek V4.1 Flash 1017DeepSeek	DeepSeek	40.0	-2	-14	high	fragile
35	Glimpseelorian	elorian	40.0	-2	-14	high	fragile
35	Xing4.0 29B A4B (free)xingchen-agi	xingchen-agi	40.0	-2	-14	high	fragile
35	Beam 1.0 501B A23Breflection	reflection	40.0	-2	-14	high	fragile
35	Mercury 3Inception	Inception	40.0	-2	-14	high	fragile
35	Ternary Bonsai 2 9Bprism-ml	prism-ml	40.0	-2	-14	high	fragile
35	Moondream 3 Previewmoondream	moondream	40.0	-2	-14	high	fragile
35	Moondream 3.1moondream	moondream	40.0	-2	-14	high	fragile
35	Muse Spark Openmeta	meta	40.0	-2	-14	high	fragile
35	Nemotron 3.5 Super VLNVIDIA	NVIDIA	40.0	-2	-14	high	fragile
35	Nemotron 3.5 Super VL (free)NVIDIA	NVIDIA	40.0	-2	-14	high	fragile
35	Pareto 26.10unbiased	unbiased	40.0	-2	-14	high	fragile
35	Pareto Latest~unbiased	~unbiased	40.0	-2	-14	high	fragile
35	Pareto 26.9unbiased	unbiased	40.0	-2	-14	high	fragile
35	LFM2.5-24B-A2BLiquid AI	Liquid AI	40.0	-2	-14	high	fragile
35	Gemini 4 ArgonGoogle	Google	40.0	-2	-14	high	fragile
35	Gemini 4 Argon (batch)Google	Google	40.0	-2	-14	high	fragile
35	Summarizescaledown	scaledown	40.0	-2	-14	high	fragile
35	Extractscaledown	scaledown	40.0	-2	-14	high	fragile
35	Muse Wondermeta	meta	40.0	-2	-14	high	fragile
35	Step 5 PreviewStepFun	StepFun	40.0	-2	-14	high	fragile
35	Ling 3.1 Flashinclusionai	inclusionai	40.0	-2	-14	high	fragile
35	GPT-6.1 Sol ProOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6.1 Sol Pro (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6.1 SolOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6.1 Sol (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	Perceptron Mk1.5perceptron	perceptron	40.0	-2	-14	high	fragile
35	Ember-1fireworks	fireworks	40.0	-2	-14	high	fragile
35	Qwen3.8 Max PrimeAlibaba	Alibaba	40.0	-2	-14	high	fragile
35	Aion 3.5 Miniaion-labs	aion-labs	40.0	-2	-14	high	fragile
35	Aion 3.5aion-labs	aion-labs	40.0	-2	-14	high	fragile
35	Solar Mini 4Upstage	Upstage	40.0	-2	-14	high	fragile
35	GPT-6 Luna ProOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6 Luna Pro (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6 LunaOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6 Luna (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6 Sol ProOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6 Sol Pro (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6 SolOpenAI	OpenAI	40.0	-2	-14	high	fragile
35	GPT-6 Sol (batch)OpenAI	OpenAI	40.0	-2	-14	high	fragile
35	MiMo-V2.6-Pro-UltraSpeedXiaomi	Xiaomi	40.0	-2	-14	high	fragile
35	MiMo-V2.6-FlashXiaomi	Xiaomi	40.0	-2	-14	high	fragile
35	MiMo-V2.6-ProXiaomi	Xiaomi	40.0	-2	-14	high	fragile
35	Qwen3.8 Omni FlashAlibaba	Alibaba	40.0	-2	-14	high	fragile
35	Ternary Bonsai 2 27Bprism-ml	prism-ml	40.0	-2	-14	high	fragile
35	DeepSeek Pro Latest~deepseek	~deepseek	40.0	-2	-14	high	fragile
35	DeepSeek Flash Latest~deepseek	~deepseek	40.0	-2	-14	high	fragile
35	Schematron V2 Turboinference-net	inference-net	40.0	-2	-14	high	fragile
35	Schematron V2 Smallinference-net	inference-net	40.0	-2	-14	high	fragile
35	GPT Astra Latest~openai	~openai	40.0	-2	-14	high	fragile
35	GPT Sol Latest~openai	~openai	40.0	-2	-14	high	fragile
35	GPT Terra Latest~openai	~openai	40.0	-2	-14	high	fragile
35	GPT Luna Latest~openai	~openai	40.0	-2	-14	high	fragile
35	Fugu Ultra v2sakana	sakana	40.0	-2	-14	high	fragile
35	Fugu Maxsakana	sakana	40.0	-2	-14	high	fragile
35	Ling 3.0 Flash VLinclusionai	inclusionai	40.0	-2	-14	high	fragile
35	DeepSeek V4.1 FlashDeepSeek	DeepSeek	40.0	-2	-14	high	fragile
35	DeepSeek V4.1 Flash (batch)DeepSeek	DeepSeek	40.0	-2	-14	high	fragile
35	Mercury 2.5Inception	Inception	40.0	-2	-14	high	fragile
35	Qwen3.8 Max (0902)Alibaba	Alibaba	40.0	-2	-14	high	fragile
35	Muse Spark 1.3 Contributormeta	meta	40.0	-2	-14	high	fragile
35	Muse Spark 1.3meta	meta	40.0	-2	-14	high	fragile
35	Granite 4.2 8BIBM	IBM	40.0	-2	-14	high	fragile
35	Hy4 previewTencent	Tencent	40.0	-2	-14	high	fragile
35	GLM Flash Latest~z-ai	~z-ai	40.0	-2	-14	high	fragile
35	Qwen3.8 FlashAlibaba	Alibaba	40.0	-2	-14	high	fragile
35	Muse Spark 1.2 Contributormeta	meta	40.0	-2	-14	high	fragile
35	DeepSeek V4 Flash Vision ExpDeepSeek	DeepSeek	40.0	-2	-14	high	fragile
35	Hy-MT2-1.8BTencent	Tencent	40.0	-2	-14	high	fragile
35	Hy-MT2-30B-A3BTencent	Tencent	40.0	-2	-14	high	fragile
35	GLM Latest~z-ai	~z-ai	40.0	-2	-14	high	fragile
35	Hy-MT2-7BTencent	Tencent	40.0	-2	-14	high	fragile
35	Qwen3.8 27BAlibaba	Alibaba	40.0	-2	-14	high	fragile
33	Mistral LargeMistral AI	Mistral AI	65.9	0	-14	high	fragile
33	Composer 2Cursor	Cursor	65.7	0	-14	high	fragile
33	Composer 2 FastCursor	Cursor	65.7	0	-14	high	fragile
33	Claude Sonnet 5.5 (batch)Anthropic	Anthropic	40.0	-2	-13	high	fragile
33	Claude Opus 5.5 (batch)Anthropic	Anthropic	40.0	-2	-13	high	fragile
31	GLM 4.5Zhipu AI	Zhipu AI	67.1	0	-13	high	fragile
31	Mistral Large 3 2512Mistral AI	Mistral AI	67.0	0	-13	high	fragile
31	Llama 3.3 70B InstructMeta	Meta	66.8	0	-13	high	fragile
31	Qwen3.5-9BAlibaba	Alibaba	66.5	0	-13	high	fragile
31	Qwen3 235B A22B Instruct 2507Alibaba	Alibaba	64.7	0	-13	high	fragile
31	Mixtral 8x22B InstructMistral AI	Mistral AI	63.4	0	-13	high	fragile
31	GPT-4o-mini (batch)OpenAI	OpenAI	62.5	0	-13	high	fragile
29	MiniMax M2.5MiniMax	MiniMax	63.6	0	-12	high	fragile
27	GLM 5Zhipu AI	Zhipu AI	68.3	0	-11	high	fragile
27	GLM 4.5VZhipu AI	Zhipu AI	67.7	0	-11	high	fragile
27	GPT-4.1OpenAI	OpenAI	67.7	0	-11	high	fragile
27	GPT-4.1 (batch)OpenAI	OpenAI	67.7	0	-11	high	fragile
27	Qwen3 235B A22B Thinking 2507Alibaba	Alibaba	64.7	0	-11	high	fragile
27	MiniMax M1MiniMax	MiniMax	64.1	0	-11	high	fragile
27	GLM 4.5 AirZhipu AI	Zhipu AI	63.7	0	-11	high	fragile
27	Phi 4Microsoft	Microsoft	61.0	0	-11	high	fragile
27	Llama 4 ScoutMeta	Meta	60.2	0	-11	high	fragile
27	GPT-4OpenAI	OpenAI	60.0	0	-11	high	fragile
27	GPT-4.1 Mini (batch)OpenAI	OpenAI	58.9	0	-11	high	fragile
27	GPT-4.1 Nano (batch)OpenAI	OpenAI	58.9	0	-11	high	fragile
25	GLM 5.3 FlashXZhipu AI	Zhipu AI	69.7	0	-10	high	fragile
25	GLM 5.3 FlashZhipu AI	Zhipu AI	69.7	0	-10	high	fragile
25	GLM 5.3 Flash (batch)Zhipu AI	Zhipu AI	69.7	0	-10	high	fragile
25	DeepSeek V3DeepSeek	DeepSeek	69.5	0	-10	high	fragile
25	Qwen3.5-27BAlibaba	Alibaba	69.3	0	-10	high	fragile
25	GPT-4o-miniOpenAI	OpenAI	69.3	0	-10	high	fragile
25	GLM 5V TurboZhipu AI	Zhipu AI	69.2	0	-10	high	fragile
25	GLM 5.3 PrimeZhipu AI	Zhipu AI	68.9	0	-10	high	fragile
25	GLM 5.3Zhipu AI	Zhipu AI	68.9	0	-10	high	fragile
25	GLM 5.3 (batch)Zhipu AI	Zhipu AI	68.9	0	-10	high	fragile
25	GLM 5.2Zhipu AI	Zhipu AI	68.9	0	-10	high	fragile
25	Qwen3.5-35B-A3BAlibaba	Alibaba	68.7	0	-10	high	fragile
25	Mistral Large 4Mistral AI	Mistral AI	68.3	0	-10	high	fragile
25	GLM 5.1Zhipu AI	Zhipu AI	68.3	0	-10	high	fragile
25	Claude Haiku 5.5 (batch)Anthropic	Anthropic	58.0	0	-10	high	fragile
25	GPT-4 Turbo (batch)OpenAI	OpenAI	57.2	0	-10	high	fragile
25	Kimi K2.5Moonshot AI	Moonshot AI	47.5	0	-10	high	fragile
25	GPT-4o-mini (2024-07-18)OpenAI	OpenAI	46.3	0	-10	high	fragile
25	GPT-4.1 MiniOpenAI	OpenAI	44.2	0	-10	high	fragile
25	Llama 3.1 8B InstructMeta	Meta	40.7	0	-10	high	fragile
25	GPT-5 MiniOpenAI	OpenAI	40.5	0	-10	high	fragile
25	Command R+ (08-2024)Cohere	Cohere	40.5	0	-10	high	fragile
23	Grok 4.20xAI	xAI	88.8	0	-9	high	fragile
23	GPT-4o (2024-08-06)OpenAI	OpenAI	71.2	0	-9	high	fragile
23	GPT-4oOpenAI	OpenAI	71.2	0	-9	high	fragile
23	GPT-4o (2024-05-13)OpenAI	OpenAI	71.2	0	-9	high	fragile
23	Gemini 3.1 Flash Lite PreviewGoogle	Google	70.7	0	-9	high	fragile
23	Qwen3.5 397B A17BAlibaba	Alibaba	70.7	0	-9	high	fragile
23	Llama 4 MaverickMeta	Meta	70.4	0	-9	high	fragile
23	Claude Haiku 4.5Anthropic	Anthropic	69.9	0	-9	high	fragile
23	Claude Haiku 4.5 (batch)Anthropic	Anthropic	69.9	0	-9	high	fragile
23	Qwen3.5-122B-A10BAlibaba	Alibaba	69.8	0	-9	high	fragile
23	Command ACohere	Cohere	51.6	0	-9	high	fragile
23	Command R (08-2024)Cohere	Cohere	40.5	+1	-9	high	fragile
21	GPT-5.1 (batch)OpenAI	OpenAI	88.7	0	-8	high	fragile
21	GPT-5.1-CodexOpenAI	OpenAI	88.7	0	-8	high	fragile
21	GPT-5.1-Codex-MiniOpenAI	OpenAI	88.7	0	-8	high	fragile
21	GPT-5 ProOpenAI	OpenAI	88.7	0	-8	high	fragile
21	GPT-5 Pro (batch)OpenAI	OpenAI	88.7	0	-8	high	fragile
21	GPT-5OpenAI	OpenAI	88.7	0	-8	high	fragile
21	GPT-5 (batch)OpenAI	OpenAI	88.7	0	-8	high	fragile
21	DeepSeek V3 0324DeepSeek	DeepSeek	71.8	0	-8	high	fragile
21	Command A+Cohere	Cohere	53.2	+1	-8	high	fragile
19	GPT-5.2 (batch)OpenAI	OpenAI	88.7	0	-7	high	fragile
19	GPT-5.1-Codex-MaxOpenAI	OpenAI	88.7	0	-7	high	fragile
19	Grok 4.3 (batch)xAI	xAI	88.5	0	-7	high	fragile
19	Gemini 3 Flash PreviewGoogle	Google	88.4	0	-7	high	fragile
19	Gemini 3 Flash Preview (batch)Google	Google	88.4	0	-7	high	fragile
17	Grok 4.20 Multi-AgentxAI	xAI	87.9	0	-6	high	fragile
17	Claude Sonnet 4.6Anthropic	Anthropic	85.2	0	-6	high	fragile
17	Claude Opus 4.5Anthropic	Anthropic	85.1	0	-6	high	fragile
17	Claude Opus 4.5 (batch)Anthropic	Anthropic	85.1	0	-6	high	fragile
17	Claude Opus 4.8 (batch)Anthropic	Anthropic	83.5	0	-6	high	fragile
17	Claude Opus 4.7 (batch)Anthropic	Anthropic	83.5	0	-6	high	fragile
17	Claude Opus 4.6 (batch)Anthropic	Anthropic	83.5	0	-6	high	fragile
17	Gemini 2.5 ProGoogle	Google	83.5	0	-6	high	fragile
17	Gemini 2.5 Pro Preview 06-05Google	Google	83.5	0	-6	high	fragile
17	DeepSeek V3.2DeepSeek	DeepSeek	83.4	0	-6	high	fragile
17	Claude Opus 4.1 (batch)Anthropic	Anthropic	82.9	0	-6	high	fragile
17	Claude Sonnet 4.5Anthropic	Anthropic	82.4	0	-6	high	fragile
17	Claude Opus 4.1Anthropic	Anthropic	82.1	0	-6	high	fragile
17	GPT-6 AstraOpenAI	OpenAI	81.8	0	-6	high	fragile
17	GPT-6 Astra (batch)OpenAI	OpenAI	81.8	0	-6	high	fragile
17	GPT-6 Astra ProOpenAI	OpenAI	81.8	0	-6	high	fragile
17	GPT-6 Astra Pro (batch)OpenAI	OpenAI	81.8	0	-6	high	fragile
17	o4 Mini (batch)OpenAI	OpenAI	81.4	0	-6	high	fragile
17	Gemini 3.8 FlashGoogle	Google	81.0	0	-6	high	fragile
17	Gemini 3.8 Flash (batch)Google	Google	81.0	0	-6	high	fragile
17	R1 0528DeepSeek	DeepSeek	79.4	0	-6	high	fragile
17	GPT-5.4 NanoOpenAI	OpenAI	79.3	0	-6	high	fragile
17	GPT-5.4 Nano (batch)OpenAI	OpenAI	79.3	0	-6	high	fragile
17	GPT-5.4 MiniOpenAI	OpenAI	79.3	0	-6	high	fragile
17	GPT-5.4 Mini (batch)OpenAI	OpenAI	79.3	0	-6	high	fragile
17	Gemini 2.5 Flash LiteGoogle	Google	79.1	0	-6	high	fragile
17	Gemini 2.5 FlashGoogle	Google	79.1	0	-6	high	fragile
17	Gemini 3.5 Flash LiteGoogle	Google	79.0	0	-6	high	fragile
17	Gemini 3.5 Flash Lite (batch)Google	Google	79.0	0	-6	high	fragile
17	Gemini 3.5 FlashGoogle	Google	79.0	0	-6	high	fragile
17	Gemini 3.5 Flash (batch)Google	Google	79.0	0	-6	high	fragile
17	Gemma 2 27BGoogle	Google	77.4	0	-6	high	fragile
17	GPT-5 Mini (batch)OpenAI	OpenAI	76.8	0	-6	high	fragile
17	GPT-5 NanoOpenAI	OpenAI	76.8	0	-6	high	fragile
17	GPT-5 Nano (batch)OpenAI	OpenAI	76.8	0	-6	high	fragile
17	Gemma 4 31BGoogle	Google	75.9	0	-6	high	fragile
17	Gemma 4 31B (free)Google	Google	75.9	0	-6	high	fragile
17	Claude Sonnet 4.6 (batch)Anthropic	Anthropic	75.3	0	-6	high	fragile
17	Claude Sonnet 4.5 (batch)Anthropic	Anthropic	75.3	0	-6	high	fragile
17	o3 Mini (batch)OpenAI	OpenAI	74.3	0	-6	high	fragile
17	GPT-4o (batch)OpenAI	OpenAI	72.0	0	-6	high	fragile
10	GPT-5.2 ProOpenAI	OpenAI	90.5	0	-5	high	stable
10	GPT-5.2 Pro (batch)OpenAI	OpenAI	90.5	0	-5	high	stable
10	GPT-5.2OpenAI	OpenAI	90.5	0	-5	high	stable
10	Claude Opus 4.6Anthropic	Anthropic	90.4	0	-5	high	stable
10	GPT-5.6 Luna ProOpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Luna Pro (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 LunaOpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Luna (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Terra ProOpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Terra Pro (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 TerraOpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Terra (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Sol ProOpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Sol Pro (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 SolOpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.6 Sol (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.5 (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
10	GPT-5.4 (batch)OpenAI	OpenAI	89.0	0	-5	high	stable
8	o3 ProOpenAI	OpenAI	86.7	0	-4	medium	stable
8	o3OpenAI	OpenAI	86.7	0	-4	medium	stable
8	o3 (batch)OpenAI	OpenAI	86.7	0	-4	medium	stable
7	Jev Routertypesafe	typesafe	40.0	-2	+191	medium	fragile
7	SwitchyardNVIDIA	NVIDIA	40.0	-2	+176	medium	fragile
6	Claude Opus 4.7Anthropic	Anthropic	95.1	0	-3	medium	stable
6	GPT-5.5 Pro (batch)OpenAI	OpenAI	92.7	0	-3	medium	stable
6	GPT-5.5OpenAI	OpenAI	92.7	0	-3	medium	stable
6	Gemini 3.1 Pro Preview Custom ToolsGoogle	Google	92.2	0	-3	medium	stable
6	Gemini 3.1 Pro PreviewGoogle	Google	92.2	0	-3	medium	stable
6	Gemini 3.1 Pro Preview (batch)Google	Google	92.2	0	-3	medium	stable
6	GPT-5.4 ProOpenAI	OpenAI	91.9	0	-3	medium	stable
6	GPT-5.4 Pro (batch)OpenAI	OpenAI	91.9	0	-3	medium	stable
6	GPT-5.4OpenAI	OpenAI	91.9	0	-3	medium	stable
5	Claude Opus 5.5Anthropic	Anthropic	95.1	0	+210	medium	fragile
5	Claude Opus 5Anthropic	Anthropic	95.1	0	+264	medium	fragile
5	GPT-5.5 ProOpenAI	OpenAI	92.7	0	+6	medium	fragile
5	GPT-5.3-CodexOpenAI	OpenAI	90.5	0	+17	medium	fragile
5	GPT-5.2-CodexOpenAI	OpenAI	90.5	0	+17	medium	fragile
5	GPT-5.2 ChatOpenAI	OpenAI	90.5	0	+35	medium	fragile
5	Grok 4.7xAI	xAI	88.8	0	+11	medium	fragile
5	Grok 4.6xAI	xAI	88.8	0	+11	medium	fragile
5	Grok 4.5xAI	xAI	88.8	0	+11	medium	fragile
5	GPT-5.1OpenAI	OpenAI	88.7	0	+8	medium	fragile
5	Claude Sonnet 5.5Anthropic	Anthropic	85.2	0	+136	medium	fragile
5	Claude Sonnet 5Anthropic	Anthropic	85.2	0	+221	medium	fragile
5	DeepSeek V3.2 ExpDeepSeek	DeepSeek	71.8	0	+253	medium	fragile
5	DeepSeek V3.1DeepSeek	DeepSeek	71.8	0	+258	medium	fragile
5	GPT-4o (2024-11-20)OpenAI	OpenAI	71.2	0	+48	medium	fragile
5	Claude Haiku 5.5Anthropic	Anthropic	69.9	0	+37	medium	fragile
5	GLM 5 TurboZhipu AI	Zhipu AI	68.3	0	+11	medium	fragile
5	GLM 4.7Zhipu AI	Zhipu AI	67.1	0	+289	medium	fragile
5	GLM 4.6Zhipu AI	Zhipu AI	67.1	0	+289	medium	fragile
5	Mistral Large 2407Mistral AI	Mistral AI	65.9	0	+254	medium	fragile
5	Kimi K2.6Moonshot AI	Moonshot AI	47.5	0	+155	medium	fragile
4	Ling 3.0 Tinyinclusionai	inclusionai	40.0	-4	+160	low	preliminary
2	Claude Opus 4.8Anthropic	Anthropic	95.1	0	-1	low	stable
2	Hy4Tencent	Tencent	40.0	-2	+262	low	preliminary

稳定模型

5 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	7d	状态
1	Claude Fable 5.1Anthropic	Anthropic	95.9	0	stable
2	Claude Fable 5.1 (batch)Anthropic	Anthropic	95.9	0	stable
3	Claude Fable 5Anthropic	Anthropic	95.9	0	stable
4	Claude Fable 5 (batch)Anthropic	Anthropic	95.9	0	stable
42	Grok 4.3xAI	xAI	88.8	+4	stable

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