参数局PARAMBUREAU

COMPARE / 2–4 MODELS

模型对比

MODEL SELECTOR

选择对比模型

4 / 4

Gemma 4 E4B (Reasoning)×Command A+×Command A×Step 3.5 Flash 2603×
Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)AnthropicGPT-6 Astra (max)OpenAIClaude Opus 5 (Adaptive Reasoning, Max Effort)AnthropicClaude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)AnthropicMuse Spark 1.3 (max)MetaGPT-5.6 Sol (max)OpenAIQwen3.8 Max (0902)阿里巴巴（通义千问）GLM-5.3 (max)智谱 AI（Z.ai）Grok 4.6 (high)SpaceXAIKimi K3 (max)月之暗面（Kimi）GPT-5.6 Terra (max)OpenAIClaude Opus 4.8 (Adaptive Reasoning, Max Effort)AnthropicGLM 5.3 Flash智谱 AI（Z.ai）Gemini 3.8 Flash (high)GoogleClaude Opus 4.7 (Adaptive Reasoning, Max Effort)AnthropicQwen3.8 Max阿里巴巴（通义千问）Qwen3.8 2.4T A95B阿里巴巴（通义千问）Qwen3.8-Flash-Next阿里巴巴（通义千问）Muse Spark 1.2 (xhigh)MetaGemini 3.7 Flash (high)GoogleDeepSeek V4.1 Flash (Reasoning, Max Effort)深度求索（DeepSeek）Grok 4.5 (high)SpaceXAIGPT-5.4 (xhigh)OpenAIGPT-5.5 (xhigh)OpenAIClaude Sonnet 5 (Adaptive Reasoning, Max Effort)AnthropicGPT-5.6 Luna (max)OpenAIDeepSeek V4 Pro 0813 (Reasoning, Max Effort)深度求索（DeepSeek）Agnes 3.0 FlashSapiens AIAgnes 2.5 Pro BetaSapiens AIDeepSeek V4 Flash 0731 Vision (Reasoning, Max Effort)深度求索（DeepSeek）DeepSeek V4 Flash 0731 (Reasoning, Max Effort)深度求索（DeepSeek）Gemini 3.6 Flash (high)GoogleMuse Spark 1.1 (xhigh)MetaGLM-5.2 (max)智谱 AI（Z.ai）Qwen3.8 27B (xhigh)阿里巴巴（通义千问）Gemini 3.5 Flash (high)GoogleMotif 3Motif TechnologiesGPT-5.3 Codex (xhigh)OpenAIMotif 3 (Beta)Motif TechnologiesClaude Opus 4.6 (Adaptive Reasoning, Max Effort)AnthropicMuse SparkMetaDeepSeek V4 Pro 0424 (Reasoning, Max Effort)深度求索（DeepSeek）K2 Horizon 375B A23BMBZUAI Institute of Foundation ModelsClaude Sonnet 4.6 (Adaptive Reasoning, Max Effort)AnthropicGPT-5.2 (xhigh)OpenAIApodex 1.1ApodexGemini 3.1 Pro PreviewGoogleQwen3.7 Max阿里巴巴（通义千问）MiniMax-M3稀宇科技（MiniMax）Claude Opus 4.5 (Reasoning)AnthropicMiMo-V2-Pro小米GPT-5.2 Codex (xhigh)OpenAIQwen3.6 Max Preview阿里巴巴（通义千问）Nex-N2-ProNex AGISolar Pro 4UpstageGemini 3 Pro Preview (high)GoogleGLM-5 (Reasoning)智谱 AI（Z.ai）Agnes 2.5 Pro AlphaSapiens AIKimi K2.6月之暗面（Kimi）JT-4.1 Flash 236B A21B中国移动Grok Build 0.1 0616SpaceXAIQuasar 438B (max, based on GLM-5.2)Multiverse ComputingQwen3.6 Plus阿里巴巴（通义千问）GPT-5.5 Instant (June 2026)OpenAIGLM-5-Turbo智谱 AI（Z.ai）GLM-5.1 (Reasoning)智谱 AI（Z.ai）MiMo-V2.5-Pro小米Gemini 3 Flash Preview (Reasoning)GoogleKimi K2.7 Code月之暗面（Kimi）Inkling SmallThinking MachinesQwen3.7 Plus阿里巴巴（通义千问）Hy3腾讯K2 Horizon MoVA 36B A4BMBZUAI Institute of Foundation ModelsGrok 4.20 0309 v2 (Reasoning)SpaceXAIInkling (xhigh)Thinking MachinesGrok 4.3 (high)SpaceXAIGrok 4.20 0309 (Reasoning)SpaceXAIMiMo-V2-Omni-0327小米Ling-3.0-flash-VL蚂蚁集团（百灵）GPT-5 Codex (high)OpenAIDeepSeek V4 Flash 0420 (Reasoning, High Effort)深度求索（DeepSeek）Solar Open2 250BUpstageGPT-5.1 (high)OpenAIGPT-5.4 mini (xhigh)OpenAIMiMo-V2-Omni小米GPT-5.1 Codex (high)OpenAIGLM 5V Turbo (Reasoning)智谱 AI（Z.ai）Kimi K2.5 (Reasoning)月之暗面（Kimi）Nemotron 3 Ultra 550B A55B (Reasoning)NVIDIAMiniMax-M2.7稀宇科技（MiniMax）Ling-3.0-flash-Fin蚂蚁集团（百灵）GPT-5 (high)OpenAIQwen3.5 27B (Reasoning)阿里巴巴（通义千问）Claude 4.1 Opus (Reasoning)AnthropicMiniMax-M2.5稀宇科技（MiniMax）Hy3-preview (Reasoning)腾讯A.X-K2SK TelecomGPT-5.5 Instant (May 2026)OpenAIGemini 3.5 Flash-LiteGoogleGrok 4SpaceXAIMiMo-V2-Flash (Feb 2026)小米MiMo-V2.5小米GLM-4.7 (Reasoning)智谱 AI（Z.ai）Kimi K2 Thinking月之暗面（Kimi）Qwen3.6 27B (Reasoning)阿里巴巴（通义千问）o3-proOpenAIG9v3-39A5BAI9StarsKAT Coder Pro V2快手（KAT）DeepSeek V3.2 (Reasoning)深度求索（DeepSeek）Qwen3.5 397B A17B (Reasoning)阿里巴巴（通义千问）Qwen3 Max Thinking阿里巴巴（通义千问）GPT-5.4 nano (xhigh)OpenAIClaude 4.5 Sonnet (Reasoning)AnthropicK2 Horizon 7BMBZUAI Institute of Foundation ModelsMiniMax-M2.1稀宇科技（MiniMax）MiMo-V2-Flash (Reasoning)小米Claude 4 Opus (Reasoning)AnthropicLing 3.0 Flash蚂蚁集团（百灵）GPT-5 mini (high)OpenAIQwen3.5 Omni Plus阿里巴巴（通义千问）GPT-5.1 Codex mini (high)OpenAIGrok 4.1 Fast (Reasoning)SpaceXAIo3OpenAIK-EXAONE 2.0 0803LG AI ResearchLongCat 2.0美团（LongCat）Step 3.7 Flash阶跃星辰（StepFun）Qwen3.5 35B A3B (Reasoning)阿里巴巴（通义千问）Claude 4 Sonnet (Reasoning)AnthropicQwen3.6 35B A3B (Reasoning)阿里巴巴（通义千问）JT-35B-Flash中国移动MiniMax-M2稀宇科技（MiniMax）KAT-Coder-Pro V1快手（KAT）GLM-4.6 (Reasoning)智谱 AI（Z.ai）Muse Glimmer (high)MetaGrok 4 Fast (Reasoning)SpaceXAIQwen3.5 122B A10B (Reasoning)阿里巴巴（通义千问）Claude 3.7 Sonnet (Reasoning)AnthropicClaude 4.5 Haiku (Reasoning)AnthropicRing-2.6-1T蚂蚁集团（百灵）Ling-2.6-1T蚂蚁集团（百灵）Step 3.5 Flash 2603阶跃星辰（StepFun）Doubao Seed Code字节跳动（豆包）Gemma 4 26B A4B (Reasoning)GoogleGemini 2.5 ProGoogleo4-mini (high)OpenAIStep 3.5 Flash阶跃星辰（StepFun）DeepSeek V3.2 Exp (Reasoning)深度求索（DeepSeek）Qwen3 Max Thinking (Preview)阿里巴巴（通义千问）K2 Horizon 3.7BMBZUAI Institute of Foundation ModelsGemini 3.1 Flash-LiteGoogleQwen3 Max阿里巴巴（通义千问）Gemini 2.5 Flash Preview (Sep '25) (Reasoning)GoogleGemma 4 31B (Reasoning)GoogleDeepSeek V3.1 Terminus (Reasoning)深度求索（DeepSeek）Kimi K2 0905月之暗面（Kimi）o1OpenAIGemini 2.5 Pro Preview (Mar' 25)GoogleMistral Medium 3.5MistralGLM-4.7-Flash (Reasoning)智谱 AI（Z.ai）Granite 4.2 30BIBMGrok 3 mini Reasoning (high)SpaceXAIGemini 2.5 Pro Preview (May' 25)GoogleDeepSeek V3.2 Speciale深度求索（DeepSeek）K-EXAONE (Reasoning)LG AI ResearchERNIE 5.0 Thinking Preview百度Gemma 4 12B (Reasoning)GoogleNova 2.0 Pro Preview (medium)AmazonGrok Code Fast 1SpaceXAICommand A+CohereApriel-v1.5-15B-ThinkerServiceNowDeepSeek V3.1 (Reasoning)深度求索（DeepSeek）Qwen3.5 9B (Reasoning)阿里巴巴（通义千问）Nova 2.0 Omni (medium)AmazonNemotron 3.5 LightningNVIDIANemotron 3 Super 120B A12B (Reasoning)NVIDIAQwen3 VL 235B A22B (Reasoning)阿里巴巴（通义千问）Apriel-v1.6-15B-ThinkerServiceNowNova 2.0 Lite (high)AmazonEXAONE 4.5 33BLG AI ResearchMiniCPM5-2B面壁智能（OpenBMB）Qwen3.5 4B (Reasoning)阿里巴巴（通义千问）DeepSeek R1 0528 (May '25)深度求索（DeepSeek）Gemini 2.5 Flash (Reasoning)GoogleGPT-5 nano (high)OpenAIGLM-4.5 (Reasoning)智谱 AI（Z.ai）Kimi K2月之暗面（Kimi）Qwen3 235B A22B 2507 (Reasoning)阿里巴巴（通义千问）GPT-4.1OpenAIQwen3 Max (Preview)阿里巴巴（通义千问）Qwen3.5 Omni Flash阿里巴巴（通义千问）o3-mini (high)OpenAIo1-proOpenAIgpt-oss-120b (high)OpenAIJT-MINI中国移动Grok 3SpaceXAISeed-OSS-36B-Instruct字节跳动（豆包）Qwen3 235B A22B 2507 Instruct阿里巴巴（通义千问）Qwen3 Coder 480B A35B Instruct阿里巴巴（通义千问）Qwen3 VL 32B (Reasoning)阿里巴巴（通义千问）Ling 3.0 Tiny蚂蚁集团（百灵）Magistral Medium 1.2MistralSonar Reasoning ProPerplexityGranite 4.2 8BIBMHyperNova 60B 2605 (high, based on gpt-oss-120b)Multiverse ComputingMiniMax M1 80k稀宇科技（MiniMax）Nemotron Cascade 2 30B A3BNVIDIAGemini 2.5 Flash Preview (Reasoning)GoogleMercury 2InceptionK2 Think V2MBZUAI Institute of Foundation ModelsLongCat Flash Lite美团（LongCat）Mistral Small 4 (Reasoning)MistralDeepSeek R1 (Jan '25)深度求索（DeepSeek）o1-previewOpenAIHyperCLOVA X SEED Think (32B)NaverGLM-4.6V (Reasoning)智谱 AI（Z.ai）Qwen3 Next 80B A3B (Reasoning)阿里巴巴（通义千问）GLM-4.5-Air智谱 AI（Z.ai）Mi:dm K 2.5 ProKorea TelecomRing-1T蚂蚁集团（百灵）Trinity Large ThinkingArcee AIG9v3-3BAI9StarsINTELLECT-3Prime IntellectGPT-5 (ChatGPT)OpenAISolar Open 100B (Reasoning)UpstageGrok 3 Reasoning BetaSpaceXAIGemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)GoogleNemotron 3 Nano Omni 30B A3B ReasoningNVIDIAGPT-4.1 miniOpenAIQwen3 Coder Next阿里巴巴（通义千问）MiniMax M1 40k稀宇科技（MiniMax）gpt-oss-20b (high)OpenAIQwen3 VL 235B A22B Instruct阿里巴巴（通义千问）North Mini CodeCohereK2-V2 (high)MBZUAI Institute of Foundation ModelsQwen3 30B A3B 2507 (Reasoning)阿里巴巴（通义千问）o1-miniOpenAIDeepSeek V3 0324深度求索（DeepSeek）Mistral Large 3MistralLing 2.6 Flash蚂蚁集团（百灵）Qwen3 Next 80B A3B Instruct阿里巴巴（通义千问）Tri-21B-think PreviewTrillion LabsQwen3 Coder 30B A3B Instruct阿里巴巴（通义千问）GPT-4.5 (Preview)OpenAIMistral Medium 3.1MistralDiffusionGemma 26B A4BGoogleQwen3 235B A22B (Reasoning)阿里巴巴（通义千问）QwQ 32B阿里巴巴（通义千问）Qwen3 VL 30B A3B (Reasoning)阿里巴巴（通义千问）Gemini 2.0 Flash Thinking Experimental (Jan '25)GoogleDevstral 2MistralLlama 4 MaverickMetaMotif-2-12.7B-ReasoningMotif TechnologiesLing-1T蚂蚁集团（百灵）Nova PremierAmazonSolar Pro 2 (Preview) (Reasoning)UpstageGranite 4.2 3BIBMMagistral Medium 1MistralMistral Medium 3MistralLlama Nemotron Super 49B v1.5 (Reasoning)NVIDIADevstral MediumMistralTri-21B-ThinkTrillion LabsGPT-4o (March 2025, chatgpt-4o-latest)OpenAIGemini 2.0 Flash (Feb '25)GoogleClaude 3.5 HaikuAnthropicLlama 3.3 Nemotron Super 49B v1 (Reasoning)NVIDIAGemma 4 E4B (Reasoning)GoogleNVIDIA Nemotron 3 Nano 30B A3B (Reasoning)NVIDIAQwen3 4B 2507 (Reasoning)阿里巴巴（通义千问）MiniCPM5-1B (Reasoning)面壁智能（OpenBMB）Sarvam 105B (high)SarvamGemini 2.0 Pro Experimental (Feb '25)GoogleDevstral Small (May '25)MistralClaude 3 OpusAnthropicSonar ReasoningPerplexityMagistral Small 1.2MistralGemini 2.5 Flash-Lite (Reasoning)GoogleDeepSeek V3 (Dec '24)深度求索（DeepSeek）GPT-4o (Nov '24)OpenAINanbeige4.1-3BNanbeigeLFM2.5-2.6BLiquid AIQwen3 VL 32B Instruct阿里巴巴（通义千问）DeepSeek R1 Distill Qwen 32B深度求索（DeepSeek）Devstral Small 2MistralMagistral Small 1MistralGemini 2.0 Flash (experimental)GoogleEXAONE 4.0 32B (Reasoning)LG AI ResearchQwen3 VL 8B (Reasoning)阿里巴巴（通义千问）DeepSeek R1 0528 Qwen3 8B深度求索（DeepSeek）Qwen2.5 Max阿里巴巴（通义千问）Qwen3 VL 30B A3B Instruct阿里巴巴（通义千问）Hermes 4 - Llama-3.1 70B (Reasoning)Nous ResearchGemini 1.5 Pro (Sep '24)GoogleDeepSeek R1 Distill Llama 70B深度求索（DeepSeek）Claude 3.5 Sonnet (Oct '24)AnthropicDeepSeek R1 Distill Qwen 14B深度求索（DeepSeek）Falcon-H1R-7BTII UAEGPT-4.1 nanoOpenAISolar Pro 3UpstageLing-flash-2.0蚂蚁集团（百灵）Gemma 4 E2B (Reasoning)GoogleQwen3 Omni 30B A3B (Reasoning)阿里巴巴（通义千问）GPT-4o (Aug '24)OpenAIQwen2.5 Instruct 72B阿里巴巴（通义千问）SonarPerplexityStep3 VL 10B阶跃星辰（StepFun）Llama 3.3 Instruct 70BMetaQwen3 30B A3B (Reasoning)阿里巴巴（通义千问）Sonar ProPerplexityDevstral Small (Jul '25)MistralQwQ 32B-Preview阿里巴巴（通义千问）GLM-4.5V (Reasoning)智谱 AI（Z.ai）Mistral Large 2 (Nov '24)MistralLlama 3.1 Nemotron Ultra 253B v1 (Reasoning)NVIDIAQwen3 30B A3B 2507 Instruct阿里巴巴（通义千问）ERNIE 4.5 300B A47B百度Hermes 4 - Llama-3.1 405B (Reasoning)Nous ResearchSolar Pro 2 (Reasoning)UpstageNVIDIA Nemotron Nano 12B v2 VL (Reasoning)NVIDIAGranite 4.1 30BIBMNVIDIA Nemotron Nano 9B V2 (Reasoning)NVIDIAGemini 2.0 Flash-Lite (Feb '25)GoogleNVIDIA Nemotron 3 Nano 4BNVIDIAMistral Small 3.1MistralQwen3 32B (Reasoning)阿里巴巴（通义千问）GPT-4o (May '24)OpenAIGemini 2.0 Flash-Lite (Preview)GoogleLlama 3.1 Nemotron Nano 4B v1.1 (Reasoning)NVIDIAKimi Linear 48B A3B Instruct月之暗面（Kimi）Llama 3.1 Instruct 405BMetaQwen3 VL 8B Instruct阿里巴巴（通义千问）Qwen3 4B (Reasoning)阿里巴巴（通义千问）LFM2.5-8B-A1BLiquid AIClaude 3.5 Sonnet (June '24)AnthropicLlama 3.1 Tulu3 405BAllen Institute for AIGPT-4o (ChatGPT)OpenAIRing-flash-2.0蚂蚁集团（百灵）Pixtral LargeMistralOlmo 3.1 32B ThinkAllen Institute for AIGrok 2 (Dec '24)SpaceXAIGemini 1.5 Flash (Sep '24)GoogleQwen3 VL 4B (Reasoning)阿里巴巴（通义千问）GPT-4 TurboOpenAIMistral Small 3.2MistralNova ProAmazonCommand ACohereQwen3.5 2B (Reasoning)阿里巴巴（通义千问）Llama 3.1 Nemotron Instruct 70BNVIDIALlama 3.1 Instruct 8BMetaGrok BetaSpaceXAIQwen2.5 Instruct 32B阿里巴巴（通义千问）Mistral Large 2 (Jul '24)MistralQwen3 4B 2507 Instruct阿里巴巴（通义千问）Qwen2.5 Coder Instruct 32B阿里巴巴（通义千问）Qwen3 14B (Reasoning)阿里巴巴（通义千问）GPT-4OpenAIMistral Small 3MistralNova LiteAmazonGPT-4o miniOpenAIDeepSeek-V2.5 (Dec '24)深度求索（DeepSeek）Llama 3.1 Instruct 70BMetaGranite 4.1 8BIBMSarvam 30B (high)SarvamGemini 2.0 Flash Thinking Experimental (Dec '24)GoogleDeepSeek-V2.5深度求索（DeepSeek）Olmo 3.1 32B InstructAllen Institute for AIMistral SabaMistralDeepSeek R1 Distill Llama 8B深度求索（DeepSeek）Olmo 3 32B ThinkAllen Institute for AILlama 4 ScoutMetaGemini 1.5 Pro (May '24)GoogleR1 1776PerplexityQwen2.5 Turbo阿里巴巴（通义千问）Reka Flash (Sep '24)Reka AILlama 3.2 Instruct 90B (Vision)MetaSolar MiniUpstageCeleris-1CelerisGrok-1SpaceXAIQwen2 Instruct 72B阿里巴巴（通义千问）Phi-4 Mini InstructMicrosoftGemini 1.5 Flash-8BGoogleQwen3.5 0.8B (Reasoning)阿里巴巴（通义千问）DeepHermes 3 - Mistral 24B Preview (Non-reasoning)Nous ResearchJamba 1.7 LargeAI21 LabsGranite 4.0 H SmallIBMMinistral 3 14BMistralJamba 1.5 LargeAI21 LabsQwen3 Omni 30B A3B Instruct阿里巴巴（通义千问）Hermes 3 - Llama-3.1 70BNous ResearchQwen3 8B (Reasoning)阿里巴巴（通义千问）DeepSeek-Coder-V2深度求索（DeepSeek）OLMo 2 32BAllen Institute for AIJamba 1.6 LargeAI21 LabsLFM2 24B A2BLiquid AIGemini 1.5 Flash (May '24)GooglePhi-4MicrosoftClaude 3 SonnetAnthropicNova MicroAmazonGranite 4.1 3BIBMMistral Small (Sep '24)MistralGemini 1.0 UltraGooglePhi-3 Mini Instruct 3.8BMicrosoftGemma 3n E4B Instruct Preview (May '25)GooglePhi-4 Multimodal InstructMicrosoftQwen2.5 Coder Instruct 7B阿里巴巴（通义千问）Mistral Large (Feb '24)MistralMixtral 8x22B InstructMistralLlama 2 Chat 7BMetaLlama 3.2 Instruct 3BMetaMiniCPM-V 4.6 1.3B面壁智能（OpenBMB）Jamba Reasoning 3BAI21 LabsQwen3 VL 4B Instruct阿里巴巴（通义千问）Qwen1.5 Chat 110B阿里巴巴（通义千问）Reka Flash 3Reka AIOlmo 3 7B ThinkAllen Institute for AIClaude 2.1AnthropicClaude 3 HaikuAnthropicOLMo 2 7BAllen Institute for AIMolmo 7B-DAllen Institute for AILing-mini-2.0蚂蚁集团（百灵）DeepSeek R1 Distill Qwen 1.5B深度求索（DeepSeek）Claude 2.0AnthropicDeepSeek-V2-Chat深度求索（DeepSeek）Mistral Small (Feb '24)MistralMistral MediumMistralGPT-3.5 TurboOpenAIMinistral 3 8BMistralLlama 3 Instruct 70BMetaArctic InstructSnowflakeQwen Chat 72B阿里巴巴（通义千问）LFM 40BLiquid AILlama 3.2 Instruct 11B (Vision)MetaPALM-2GoogleGemini 1.0 ProGoogleDeepSeek Coder V2 Lite Instruct深度求索（DeepSeek）Sarvam M (Reasoning)SarvamDeepSeek LLM 67B Chat (V1)深度求索（DeepSeek）Llama 2 Chat 70BMetaLlama 2 Chat 13BMetaCommand-R+ (Apr '24)CohereOpenChat 3.5 (1210)OpenChatDBRX InstructDatabricksExaone 4.0 1.2B (Reasoning)LG AI ResearchOlmo 3 7B InstructAllen Institute for AILFM2.5-1.2B-ThinkingLiquid AIJamba 1.7 MiniAI21 LabsLFM2 2.6BLiquid AILFM2.5-1.2B-InstructLiquid AIJamba 1.5 MiniAI21 LabsGranite 4.0 H 1BIBMQwen3 1.7B (Reasoning)阿里巴巴（通义千问）Jamba 1.6 MiniAI21 LabsMixtral 8x7B InstructMistralGemma 3 270MGoogleApertus 70B InstructSwiss AI InitiativeGranite 4.0 MicroIBMDeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)Nous ResearchClaude InstantAnthropicCommand-R (Mar '24)CohereLlama 65BMetaMistral 7B InstructMistralQwen Chat 14B阿里巴巴（通义千问）Granite 4.0 1BIBMMolmo2-8BAllen Institute for AILFM2 8B A1BLiquid AIGranite 3.3 8B (Non-reasoning)IBMGemma 3 27B InstructGoogleMinistral 3 3BMistralApertus 8B InstructSwiss AI InitiativeGemma 3 1B InstructGoogleGemma 3 4B InstructGoogleGemma 3n E2B InstructGoogleGemma 3n E4B InstructGoogleGranite 4.0 350MIBMGranite 4.0 H 350MIBMLFM2 1.2BLiquid AILFM2.5-VL-1.6BLiquid AILlama 3 Instruct 8BMetaLlama 3.2 Instruct 1BMetaQwen3 0.6B (Reasoning)阿里巴巴（通义千问）Tiny Aya GlobalCohereGemma 3 12B InstructGoogleK2 Horizon 0.9BMBZUAI Institute of Foundation ModelsCogito v2.1 (Reasoning)Deep CogitoGemini 3 Deep ThinkGoogleGPT-3.5 Turbo (0613)OpenAIGPT-4o mini Realtime (Dec '24)OpenAIGPT-4o Realtime (Dec '24)OpenAIGPT-5.4 Pro (xhigh)OpenAIGPT-5.5 Pro (xhigh)OpenAIMi:dm K 2.5 Pro PreviewKorea Telecom

Gemma 4 E4B (Reasoning) / Command A+ / Command A / Step 3.5 Flash 2603

PRICE / CAPABILITY FRONTIER

模型斩杀线坐标

4 个所选模型

左上区域代表更低价格和更高能力。红色前沿线由当前主流模型计算；对比状态只显示所选模型点，隐藏其他背景点。价格采用标准输入档位，横轴为对数刻度。

020406080100

AA 智能指数越高越好

Gemma 4 E4B (Reasoning)Google
8.915差 47.4%
Command A+Cohere
13.924差 17.9%
Command ACohere
6.958差 59.0%
Step 3.5 Flash 2603StepFun
16.964最优

GPQA 科学推理越高越好

Gemma 4 E4B (Reasoning)Google
0.576差 30.3%
Command A+Cohere
0.761差 7.9%
Command ACohere
0.527差 36.2%
Step 3.5 Flash 2603StepFun
0.826最优

τ² 智能体工具任务越高越好

Gemma 4 E4B (Reasoning)Google
0.208差 76.3%
Command A+Cohere
0.807差 7.7%
Command ACohere
0.152差 82.6%
Step 3.5 Flash 2603StepFun
0.874最优

SciCode 科研编程越高越好

Gemma 4 E4B (Reasoning)Google
—
Command A+Cohere
0.385最优
Command ACohere
—
Step 3.5 Flash 2603StepFun
—

输出速度中位数越高越好

Gemma 4 E4B (Reasoning)Google
42.786差 82.7%
Command A+Cohere
246.96最优
Command ACohere
53.242差 78.4%
Step 3.5 Flash 2603StepFun
122.783差 50.3%

条长为同一维度内的 0至100 归一化分数；右侧数字为来源原始值。指数使用原始百分制，速度和延迟按本次所选模型相对归一化。

评测数据来自 Artificial Analysis，本站仅作非官方中文呈现；归一化仅用于当前所选模型的视觉比较，不改变原始值。查看指标口径。

FULL SPECIFICATION

完整指标对照

35 项

指标Gemma 4 E4B (Reasoning)Command A+Command AStep 3.5 Flash 2603
AA 智能指数 · 越高越好8.915距最优 47.4%13.924距最优 17.9%6.958距最优 59.0%16.964最优
智能指数评测成本USD · 越低越好—无数据0最优—无数据—无数据
CritPT 批判性推理 · 越高越好0.006距最优 75.0%0.003距最优 87.5%0距最优 100.0%0.023最优
GDPval 专业任务 · 越高越好0距最优 100.0%0.079最优—无数据—无数据
GPQA 科学推理 · 越高越好0.576距最优 30.3%0.761距最优 7.9%0.527距最优 36.2%0.826最优
Humanity’s Last Exam · 越高越好0.038距最优 84.7%0.12距最优 51.1%0.04距最优 83.5%0.245最优
IFBench 指令遵循 · 越高越好0.442距最优 40.2%0.74最优0.365距最优 50.7%0.665距最优 10.0%
LCR 长上下文推理 · 越高越好0.32距最优 49.2%0.527距最优 16.4%0.213距最优 66.1%0.63最优
MMMU-Pro 多模态理解 · 越高越好0.514距最优 18.7%0.632最优—无数据—无数据
Omniscience 综合知识 · 越高越好-19.7距最优 390.5%-4.017最优-48.267距最优 1101.7%-44.217距最优 1000.8%
Omniscience 准确率 · 越高越好0.086距最优 65.1%0.089距最优 63.9%0.164距最优 33.4%0.246最优
Omniscience 非幻觉率 · 越高越好0.691距最优 19.5%0.858最优0.227距最优 73.6%0.087距最优 89.9%
SciCode 科研编程 · 越高越好—无数据0.385最优—无数据—无数据
τ² 智能体工具任务 · 越高越好0.208距最优 76.3%0.807距最优 7.7%0.152距最优 82.6%0.874最优
τ-Bench 银行业任务 · 越高越好—无数据0.06最优—无数据—无数据
Terminal-Bench Hard · 越高越好0.083距最优 74.4%0.25距最优 23.3%0.008距最优 97.7%0.326最优
输出速度中位数tokens/s · 越高越好42.786距最优 82.7%246.96最优53.242距最优 78.4%122.783距最优 50.3%
输出速度 P05tokens/s · 越高越好21.396距最优 87.8%175.728最优23.002距最优 86.9%82.871距最优 52.8%
输出速度 P25tokens/s · 越高越好36.763距最优 83.6%223.794最优35.059距最优 84.3%112.598距最优 49.7%
输出速度 P75tokens/s · 越高越好48.179距最优 82.1%268.83最优64.2距最优 76.1%148.172距最优 44.9%
输出速度 P95tokens/s · 越高越好54.299距最优 81.8%298.295最优72.751距最优 75.6%194.91距最优 34.7%
推理耗时中位数秒 · 越低越好46.744距最优 477.2%8.099最优—无数据16.289距最优 101.1%
首 token 延迟中位数秒 · 越低越好0.865距最优 112.0%0.408最优1.82距最优 346.1%3.411距最优 736.0%
首 token 延迟 P05秒 · 越低越好0.735距最优 117.4%0.338最优1.544距最优 356.8%2.924距最优 764.7%
首 token 延迟 P25秒 · 越低越好0.782距最优 108.1%0.376最优1.637距最优 335.6%3.235距最优 760.7%
首 token 延迟 P75秒 · 越低越好0.982距最优 128.9%0.429最优2.371距最优 452.7%3.59距最优 736.7%
首 token 延迟 P95秒 · 越低越好1.205距最优 47.4%0.817最优3.012距最优 268.7%7.205距最优 781.8%
首答案 token 延迟中位数秒 · 越低越好47.609距最优 2515.6%8.507距最优 367.3%1.82最优19.7距最优 982.3%
端到端响应中位数秒 · 越低越好59.296距最优 463.0%10.531最优11.211距最优 6.5%23.772距最优 125.7%
上下文窗口tokens · 越高越好128,000距最优 50.0%192,000距最优 25.0%256,000最优256,000最优
Standard 输入价格百万 tokens · 越低越好¥0.1341不可计算¥0.0000最优¥16.7678不可计算¥0.6707不可计算
Standard 输出价格百万 tokens · 越低越好¥0.6707不可计算¥0.0000最优¥67.0712不可计算¥2.0121不可计算
Standard 缓存输入价格百万 tokens · 越低越好—无数据—无数据—无数据¥0.1341最优
terminal bench21 · 越高越好0.019距最优 91.8%0.229最优—无数据—无数据
terminal bench40 · 越高越好—无数据0.005最优—无数据—无数据

人民币价格按服务端汇率 1 USD = 6.7071 CNY 折算，参考日期 2026/9/16。