Economics & Performance

成本、速度与响应

把模型能力放回真实调用条件中,分别观察成本、任务耗时、回答生成速度和回答长度。

数据快照:2026-09-19查看原始来源 ↗
越低越好

单次 Intelligence 任务成本

US$0.01

MiMo-V2.5 / Llama 4 Scout

v4.1 历史参考榜单中最低的非零公开任务成本;不代表当前 API 定价,也不与 v4.3 智能指数组合排名。

查看原始数据 ↗
越低越好

单次 Intelligence 任务耗时

1.6 min

Qwen3.7 Max

按各评测权重计算的平均解码时间,不含 TTFT。

查看原始数据 ↗
越高越好

回答生成速度

2,038 Token/s

Celeris-1

模型开始回答后,每秒生成的文本单位数量;数值越高,回答越快。

查看原始数据 ↗
结合任务理解

单次任务回答长度

按 Token 统计

按模型比较

Token 是模型处理文本的基本单位,大致对应一个字、词或词片段。

查看原始数据 ↗

FULL PERFORMANCE DATA

完整成本与响应数据

覆盖当前模型总榜的全部 289 条记录;“—”表示来源快照没有公开该项。

151 / 289 个模型
模型智能、价格、速度和延迟横向比较 · v4.1 · 历史参考
模型上下文机构USDToken/s进一步分析
Claude Opus 5 (max)推理模型Anthropic · 1M Token1MAnthropic61US$2.035580.0189.14
Grok 4.6 (high)推理模型SpaceXAI · 500k Token500kSpaceXAI61US$0.846539.5747.21
Claude Opus 5 (xhigh)推理模型Anthropic · 1M Token1MAnthropic60US$1.565442.9352.21
Claude Fable 5推理模型Anthropic · 1M Token1MAnthropic60US$2.7573197.59204.43
GLM-5.3 (max)推理模型Z AI · 1M Token1MZ AI59.51US$0.6874.0172.05335.829
GPT-5.6 Sol (max)推理模型OpenAI · 1M Token1MOpenAI59US$1.5481144.31150.51
Claude Opus 5 (high)推理模型Anthropic · 1M Token1MAnthropic59US$1.065520.930.02
GPT-5.6 Sol (xhigh)推理模型OpenAI · 1M Token1MOpenAI58US$0.958238.1144.23
Qwen3.8 Max推理模型Alibaba · 1M Token1MAlibaba58US$1.13472.6355.65
Qwen3.8 2.4T A95B推理模型Alibaba · 984k Token984kAlibaba58US$1.09502.7552.65
Kimi K3推理模型Kimi · 1.05M Token1.05MKimi57US$0.7233144.34219.46
Muse Spark 1.2 (xhigh)推理模型Meta · 1.05M Token1.05MMeta57US$0.40
Claude Opus 5 (medium)推理模型Anthropic · 1M Token1MAnthropic56US$0.62535.7815.25
GPT-5.6 Sol (high)推理模型OpenAI · 1M Token1MOpenAI56US$0.627812.4818.92
Claude Opus 4.8 (max)推理模型Anthropic · 1M Token1MAnthropic56US$1.805630.9539.81
Gemini 3.7 Flash (high)推理模型Google · 1M Token1MGoogle56US$0.403409.8311.3
GPT-5.6 Terra (max)推理模型OpenAI · 1M Token1MOpenAI55US$0.82137177.83181.47
Grok 4.5 (high)推理模型SpaceXAI · 500k Token500kSpaceXAI54US$0.35609.7618.04
GPT-5.6 Sol (medium)推理模型OpenAI · 1M Token1MOpenAI54US$0.41744.4811.23
Claude Sonnet 5 (max)推理模型Anthropic · 1M Token1MAnthropic53US$1.5374161167.79
Gemini 3.7 Flash (medium)推理模型Google · 1M Token1MGoogle53US$0.262744.226.05
DeepSeek V4 Pro 0813 (max)推理模型DeepSeek · 1M Token1MDeepSeek53US$0.25801.8233.07
Qwen3.8 27B推理模型Alibaba · 256k Token256kAlibaba52.02
GPT-5.6 Terra (xhigh)推理模型OpenAI · 1M Token1MOpenAI52US$0.4813721.6625.32
DeepSeek V4 Flash 0731推理模型DeepSeek · 1M Token1MDeepSeek52US$0.031201.4322.27
GPT-5.6 Luna (max)推理模型OpenAI · 1M Token1MOpenAI51US$0.28198139.79142.31
GLM-5.2 (max)推理模型Z AI · 1M Token1MZ AI51US$0.322201.3212.7
Muse Spark 1.1 (xhigh)推理模型Meta · 1.05M Token1.05MMeta51US$0.261291.4620.88
Claude Opus 5 (low)推理模型Anthropic · 1M Token1MAnthropic51US$0.36522.9812.68
Gemini 3.7 Flash (low)推理模型Google · 1M Token1MGoogle51US$0.162530.742.71
Gemini 3.5 Flash推理模型Google · 1M Token1MGoogle50US$0.5918824.8827.54
Gemini 3.6 Flash推理模型Google · 1M Token1MGoogle50US$0.5024018.3820.46
GPT-5.6 Sol (low)推理模型OpenAI · 1M Token1MOpenAI49US$0.24763.099.7
GPT-5.6 Luna (xhigh)推理模型OpenAI · 1M Token1MOpenAI49US$0.1818435.6538.36
GPT-5.6 Terra (high)推理模型OpenAI · 1M Token1MOpenAI49US$0.341222.326.41
Kimi K3 (low)推理模型Kimi · 1.05M Token1.05MKimi47US$0.24334.7780.13
Motif 3推理模型Motif Technologies · 262k Token262kMotif Technologies47
Gemini 3.1 Pro Preview推理模型Google · 1M Token1MGoogle46US$0.2912941.2745.13
GPT-5.6 Luna (high)推理模型OpenAI · 1M Token1MOpenAI46US$0.121979.0711.6
Qwen3.7 Max推理模型Alibaba · 1M Token1MAlibaba46US$1.032022.6117.02
GPT-5.6 Terra (medium)推理模型OpenAI · 1M Token1MOpenAI46US$0.181301.645.49
MiniMax-M3推理模型MiniMax · 1M Token1MMiniMax44US$0.12891.6329.64
DeepSeek V4 Pro (max)推理模型DeepSeek · 1M Token1MDeepSeek44US$0.04731.5168.11
Motif 3 (Beta)推理模型Motif Technologies · 262k Token262kMotif Technologies44
DeepSeek V4 Pro (high)推理模型DeepSeek · 1M Token1MDeepSeek43US$0.04751.6234.8
Muse Spark推理模型Meta · 262k Token262kMeta43
MiMo-V2.5-Pro推理模型Xiaomi · 1M Token1MXiaomi42US$0.06712.9638.41
Kimi K2.7 Code推理模型Kimi · 256k Token256kKimi42452.8263.96
Claude Sonnet 5 (Non-reasoning)非推理模型Anthropic · 1M Token1MAnthropic42US$0.37621.219.31
Solar Pro 4推理模型Upstage · 512k Token512kUpstage42US$0.22652.240.53
Hy3推理模型Tencent · 256k Token256kTencent41US$0.03602.4944.33
GPT-5.6 Sol (Non-reasoning)非推理模型OpenAI · 1M Token1MOpenAI41US$0.267517.63
Nex-N2-Pro推理模型Nex AGI · 262k Token262kNex AGI411361.7620.18
Inkling推理模型Thinking Machines · 1M Token1MThinking Machines41871.7330.33
GPT-5.6 Terra (low)推理模型OpenAI · 1M Token1MOpenAI40US$0.151291.385.25
DeepSeek V4 Flash 0420推理模型DeepSeek · 1M Token1MDeepSeek40US$0.021181.2552.98
Qwen3.6 Plus推理模型Alibaba · 1M Token1MAlibaba40US$0.31522.51119.07
Inkling Small推理模型Thinking Machines · 1M Token1MThinking Machines40US$0.07951.6727.96
Qwen3.7 Plus推理模型Alibaba · 1M Token1MAlibaba39US$0.22522.8750.64
JT-4.1 Flash 236B A21B推理模型China Mobile · 256k Token256kChina Mobile39
Agnes 2.5 Pro Alpha推理模型Sapiens AI · 1M Token1MSapiens AI391031.6625.83
GPT-5.6 Luna (medium)推理模型OpenAI · 1M Token1MOpenAI38US$0.071962.845.39
Nemotron 3 Ultra推理模型NVIDIA · 262k Token262kNVIDIA38US$0.252001.1415.02
Ling 3.0 Flash推理模型InclusionAI · 262k Token262kInclusionAI38US$0.043822.028.57
DeepSeek V4 Flash 0420 · AA High Effort推理模型DeepSeek · 1M Token1MDeepSeek37US$0.04
MiMo-V2.5推理模型Xiaomi · 1M Token1MXiaomi37US$0.01694.3940.65
Qwen3.6 27B推理模型Alibaba · 262k Token262kAlibaba37US$0.27563.66114.28
Solar Open2 250B推理模型Upstage · 1.05M Token1.05MUpstage37
Gemini 3.5 Flash-Lite推理模型Google · 1M Token1MGoogle36US$0.094288.119.28
Muse Glimmer (high)推理模型Meta · 131k Token131kMeta35US$0.07990.8326.01
A.X-K2推理模型SK Telecom · 262k Token262kSK Telecom35
GLM-5.2推理模型Z AI · 1M Token1MZ AI341221.545.62
GPT-5.6 Terra (Non-reasoning)非推理模型OpenAI · 1M Token1MOpenAI34US$0.181330.784.54
KAT-Coder-Pro V2推理模型KwaiKAT · 256k Token256kKwaiKAT341031.576.42
Qwen3.5 397B A17B推理模型Alibaba · 262k Token262kAlibaba34US$0.33702.3655.13
LongCat 2.0推理模型LongCat · 1M Token1MLongCat33
GPT-5.6 Luna (low)推理模型OpenAI · 1M Token1MOpenAI33US$0.061741.554.42
Qwen3.5 122B A10B推理模型Alibaba · 262k Token262kAlibaba32US$0.241332.3221.09
Qwen3.6 35B A3B推理模型Alibaba · 262k Token262kAlibaba32US$0.181582.2239.49
G9v3-39A5B推理模型AI9Stars · 131k Token131kAI9Stars32*
Ring-2.6-1T推理模型InclusionAI · 262k Token262kInclusionAI31US$0.351263.3923.3
K-EXAONE 2.0推理模型LG AI Research · 262k Token262kLG AI Research31
Qwen3.6 27B推理模型Alibaba · 262k Token262kAlibaba30US$0.36573.6312.4
Step 3.7 Flash推理模型StepFun · 262k Token262kStepFun30US$0.094050.927.09
Mistral Medium 3.5推理模型Mistral · 256k Token256kMistral30US$0.39104226.11
Claude 4.5 Haiku推理模型Anthropic · 200k Token200kAnthropic30US$0.2410614.4619.17
Gemma 4 31B推理模型Google · 256k Token256kGoogle29351.1265.8
GPT-5.5 Instant (June 2026)推理模型OpenAI · 400k Token400kOpenAI29US$0.54
Qwen3.5 122B A10B推理模型Alibaba · 262k Token262kAlibaba28US$0.181462.375.79
GPT-5.6 Luna (Non-reasoning)非推理模型OpenAI · 1M Token1MOpenAI27US$0.081790.793.58
Gemini 2.5 Pro推理模型Google · 1M Token1MGoogle26US$0.2014322.2625.75
Gemma 4 26B A4B推理模型Google · 256k Token256kGoogle26US$0.03
NVIDIA Nemotron 3 Super推理模型NVIDIA · 1M Token1MNVIDIA25US$0.211871.6515
Gemini 3.1 Flash-Lite推理模型Google · 1M Token1MGoogle25US$0.043296.017.53
Grok 4.3 (Non-reasoning)非推理模型SpaceXAI · 1M Token1MSpaceXAI25US$0.371010.825.75
MiMo-V2-Flash推理模型Xiaomi · 256k Token256kXiaomi25
Ling 3.0 Tiny推理模型InclusionAI · 262k Token262kInclusionAI25US$0.002032.4514.76
Qwen3.6 35B A3B推理模型Alibaba · 262k Token262kAlibaba24US$0.602002.284.78
Qwen3.5 35B A3B推理模型Alibaba · 262k Token262kAlibaba24US$0.231342.25.93
gpt-oss-120b (high)推理模型OpenAI · 131k Token131kOpenAI24US$0.062840.859.66
Nemotron 3.5 Lightning推理模型NVIDIA · 1M Token1MNVIDIA24US$0.062931.129.64
Command A+推理模型Cohere · 192k Token192kCohere23US$0.002030.412.7
K-EXAONE推理模型LG AI Research · 256k Token256kLG AI Research22
Gemma 4 12B推理模型Google · 256k Token256kGoogle221082.4425.5
Gemma 4 31B推理模型Google · 256k Token256kGoogle22US$0.03732.289.1
Nova 2.0 Pro Preview (medium)推理模型Amazon · 256k Token256kAmazon22US$0.1711715.7136.99
Qwen3.5 9B推理模型Alibaba · 262k Token262kAlibaba21US$0.22751.5735.04
Mercury 2推理模型Inception · 128k Token128kInception21US$0.089873.934.43
Qwen3 Coder Next推理模型Alibaba · 256k Token256kAlibaba21US$0.331311.275.09
EXAONE 4.5 33B推理模型LG AI Research · 262k Token262kLG AI Research20
North Mini Code推理模型Cohere · 256k Token256kCohere20US$0.00640.4839.46
Nova 2.0 Pro Preview (low)推理模型Amazon · 256k Token256kAmazon20US$0.211219.730.38
Mistral Small 4推理模型Mistral · 256k Token256kMistral20US$0.101630.7516.08
Devstral 2推理模型Mistral · 256k Token256kMistral19US$0.00361.2515.28
Nova 2.0 Lite (high)推理模型Amazon · 1M Token1MAmazon18US$0.2514918.9235.71
Trinity Large Thinking推理模型Arcee AI · 512k Token512kArcee AI18US$0.131660.915.97
Magistral Medium 1.2推理模型Mistral · 128k Token128kMistral18US$0.75441.7158.39
HyperNova 60B 2605推理模型Multiverse Computing · 131k Token131kMultiverse Computing18US$0.023640.917.78
Nemotron Cascade 2 30B A3B推理模型NVIDIA · 1M Token1MNVIDIA18
Devstral Small 2推理模型Mistral · 256k Token256kMistral17US$0.00341.3316.02
K2 Think V2推理模型MBZUAI Institute of Foundation Models · 262k Token262kMBZUAI Institute of Foundation Models17
Qwen3 Next 80B A3B推理模型Alibaba · 262k Token262kAlibaba17US$0.172102.2414.12
G9v3-3B推理模型AI9Stars · 131k Token131kAI9Stars16
Mistral Large 3推理模型Mistral · 256k Token256kMistral16US$0.06491.1411.4
gpt-oss-120b (low)推理模型OpenAI · 131k Token131kOpenAI15US$0.022950.859.32
gpt-oss-20b (high)推理模型OpenAI · 131k Token131kOpenAI15US$0.021970.8213.53
Nova 2.0 Pro Preview推理模型Amazon · 256k Token256kAmazon14US$0.251091.035.6
Llama 4 Maverick推理模型Meta · 1M Token1MMeta14US$0.031050.935.7
NVIDIA Nemotron 3 Nano推理模型NVIDIA · 1M Token1MNVIDIA14US$0.021561.3717.44
Solar Pro 3推理模型Upstage · 128k Token128kUpstage14
Ling 2.6 Flash推理模型InclusionAI · 262k Token262kInclusionAI141301.134.97
DiffusionGemma 26B A4B推理模型Google · 256k Token256kGoogle13
Gemma 4 E4B推理模型Google · 128k Token128kGoogle12900.7928.67
Celeris-1推理模型Celeris · 131k Token131kCeleris122,0380.650.9
Magistral Small 1.2推理模型Mistral · 128k Token128kMistral11US$0.25880.9329.29
Nanbeige4.1-3B推理模型Nanbeige · 256k Token256kNanbeige11
Ministral 3 14B推理模型Mistral · 256k Token256kMistral11US$0.15620.858.9
Llama 4 Scout推理模型Meta · 10M Token10MMeta10US$0.01960.765.98
Gemma 4 E2B推理模型Google · 128k Token128kGoogle9
Llama 3.3 70B推理模型Meta · 128k Token128kMeta9US$0.08851.667.54
Ministral 3 8B推理模型Mistral · 256k Token256kMistral9US$0.181120.735.18
Granite 4.1 30B推理模型IBM · 131k Token131kIBM9
NVIDIA Nemotron 3 Nano 4B推理模型NVIDIA · 262k Token262kNVIDIA9
Qwen3.5 2B推理模型Alibaba · 262k Token262kAlibaba7
Ministral 3 3B推理模型Mistral · 256k Token256kMistral6US$0.132510.632.62
Phi-4 Mini推理模型Microsoft · 128k Token128kMicrosoft6US$0.00450.8311.96
Qwen3.5 2B推理模型Alibaba · 262k Token262kAlibaba6
Qwen3.5 0.8B推理模型Alibaba · 262k Token262kAlibaba5
Granite 4.1 3B推理模型IBM · 131k Token131kIBM5
MiniCPM-V 4.6 1.3B推理模型OpenBMB · 262k Token262kOpenBMB4
Qwen3.5 0.8B推理模型Alibaba · 262k Token262kAlibaba3
已显示 12 / 151 个模型

仅比较同一评测版本;v4.1 保留为历史参考,不与 v4.3 混排。单任务成本不是当前 API 价格。“—”表示该项未知;带 * 的指数来自原始榜单的估算标记。

OUTPUT TOKENS

单次 Intelligence 任务输出 Token

直接展示来源当前公开的 20 条回答 / 推理 Token 数据,不截成首页预览。

回答与推理 Token 拆分
模型回答 Token推理 Token合计
GPT-5.6 Sol (max)5,27010,07615,346
GPT-5.6 Sol (xhigh)4,3925,5499,941
Kimi K35,79617,83723,632
GPT-5.6 Sol (high)3,5033,1876,690
GPT-5.6 Terra (max)5,18614,18419,370
Grok 4.5 (high)6,3047,52613,830
GPT-5.6 Sol (medium)2,6781,5244,203
GPT-5.6 Terra (xhigh)4,2106,82511,036
GPT-5.6 Luna (max)5,00613,90618,912
Muse Spark 1.1 (xhigh)6,47615,13521,610
Gemini 3.5 Flash9,81917,91427,733
GPT-5.6 Sol (low)1,8916162,508
GPT-5.6 Luna (xhigh)4,0778,41612,492
GPT-5.6 Terra (high)3,5104,2287,738
MiniMax-M311,62312,34023,964
GPT-5.5 (xhigh)3,57212,49216,064
Claude Opus 4.7 (max)9,75721,70931,466
GPT-5.5 (high)3,1446,93310,076
GPT-5.5 (medium)2,5132,6205,132
Nemotron 3 Ultra7,79715,14122,938

COST PER TASK

单次任务成本组成

回答、推理、缓存写入、缓存命中和普通输入分别计价。

每项 Intelligence Index 任务的加权平均成本(USD)
模型回答推理缓存写入缓存命中输入合计
GPT-5.6 Sol (xhigh)0.1320.1660.1510.2000.032US$0.68
Kimi K30.0870.2680.4040.1760.020US$0.95
GPT-5.6 Sol (high)0.1050.0960.0900.1300.032US$0.45
GPT-5.6 Terra (max)0.0780.2130.3730.1420.019US$0.82
Grok 4.5 (high)0.0380.0450.0610.1560.013US$0.31
GPT-5.6 Sol (medium)0.0800.0460.0850.0710.032US$0.31
GPT-5.6 Terra (xhigh)0.0630.1020.2170.0790.016US$0.48
GPT-5.6 Luna (max)0.0300.0830.0260.0620.008US$0.21
GLM-5.2 (max)0.0270.1610.1720.0980.010US$0.47
Muse Spark 1.1 (xhigh)0.0280.0640.0720.0890.008US$0.26
Gemini 3.5 Flash0.0880.1610.2040.1210.011US$0.59
GPT-5.6 Sol (low)0.0570.0180.0560.0340.032US$0.20
GPT-5.6 Luna (xhigh)0.0240.0500.0190.0380.007US$0.14
GPT-5.6 Terra (high)0.0530.0630.1520.0530.016US$0.34
MiniMax-M30.0140.0150.0230.0700.002US$0.12
DeepSeek V4 Pro (max)0.0060.0260.0090.0010.003US$0.04
gpt-oss-120b (high)0.0050.0170.0110.0270.001US$0.06
GPT-5.5 (high)0.0940.2080.2550.0790.032US$0.67
GPT-5.5 (medium)0.0750.0790.1720.0480.032US$0.41
Nemotron 3 Ultra0.0210.0410.0990.0790.005US$0.24

API PRICING

每百万 Token 公开价格

价格与缓存折扣会随 Provider 更新;数字必须连同 2026-09-19 快照日期阅读。

Cache Hit、输入和输出价格(USD / 1M Token)
模型缓存命中输入输出
GPT-5.6 Sol (max)US$0.5US$5US$30
Kimi K3US$0.3US$3US$15
Claude Opus 4.8 (max)US$0.5US$5US$25
GPT-5.6 Terra (max)US$0.25US$2.5US$15
Grok 4.5 (high)US$0.5US$2US$6
Claude Sonnet 5 (max)US$0.2US$2US$10
GPT-5.6 Terra (xhigh)US$0.25US$2.5US$15
GPT-5.6 Luna (max)US$0.1US$1US$6
GLM-5.2 (max)US$0.26US$1.4US$4.4
Muse Spark 1.1 (xhigh)US$0.15US$1.25US$4.25
Gemini 3.5 FlashUS$0.15US$1.5US$9
GPT-5.6 Luna (xhigh)US$0.1US$1US$6
GPT-5.6 Terra (high)US$0.25US$2.5US$15
MiniMax-M3US$0.06US$0.3US$1.2
DeepSeek V4 Pro (max)US$0.0036US$0.435US$0.87
gpt-oss-120b (high)US$0.15US$0.15US$0.6
Claude Opus 4.7 (max)US$0.5US$5US$25
GPT-5.4 (xhigh)US$0.25US$2.5US$15
Claude Sonnet 4.6 (max)US$0.3US$3US$15
Nemotron 3 UltraUS$0.25US$0.675US$2.675

TIME PER TASK

单次 Intelligence 任务加权耗时

按评测权重计算的平均解码时间,不包含首字等待。

当前公开的 20 条任务耗时
模型分钟 / 任务
GPT-5.6 Sol (xhigh)2.71 分钟
GPT-5.6 Sol (high)1.83 分钟
GPT-5.6 Terra (max)2.06 分钟
Grok 4.5 (high)3.27 分钟
GPT-5.6 Sol (medium)1.23 分钟
GPT-5.6 Terra (xhigh)1.35 分钟
GPT-5.6 Luna (max)1.57 分钟
GLM-5.2 (max)3.29 分钟
Muse Spark 1.1 (xhigh)2.77 分钟
Gemini 3.5 Flash2.69 分钟
GPT-5.6 Sol (low)0.77 分钟
GPT-5.6 Luna (xhigh)1.05 分钟
GPT-5.6 Terra (high)0.94 分钟
MiniMax-M33.87 分钟
gpt-oss-120b (high)1.87 分钟
GPT-5.5 (xhigh)3.37 分钟
GPT-5.5 (high)2.16 分钟
GPT-5.4 (xhigh)3.52 分钟
GPT-5.5 (medium)1.08 分钟
Nemotron 3 Ultra1.79 分钟

为什么不能只看一个指标

高生成速度不代表首字等待短,低单价也不代表完成同一任务更便宜。一个模型可能输出更长、推理更久,最终任务成本和完整响应时间仍然更高。