输出速度榜
按 Artificial Analysis · 输出速度实测 口径排列的 26 款大模型,Celeris-1、Mercury 2.5、Muse Ember Max 领跑榜单。数据聚合自公开评测源,更新于 2026-11-21。
口径:按实测输出速度排序,实时对话与高吞吐场景的核心口径。,分数越高越好。评测任务成本不是 API Token 单价;不同来源的分数不能直接相加或横向比较。
输出速度榜
| 排名 | 模型 | 输出速度 tok/s | 上下文 | 输入 /M | 输出 /M | 详情 |
|---|---|---|---|---|---|---|
| 1 | Celeris-1Inception Labs | 722 tok/s | 128K | $0.70 | $2.10 | |
| 2 | Mercury 2.5Inception Labs | 531 tok/s | 128K | $0.50 | $1.50 | |
| 3 | Muse Ember MaxMeta | 463 tok/s | 256K | $0.45 | $1.80 | |
| 4 | Claude Haiku 5.5Anthropic | 240 tok/s | 1M | $1 | $5 | |
| 5 | Ling 3.1 Flash蚂蚁集团 | 214 tok/s | 1M | $0.40 | $1.60 | |
| 6 | Muse Spark 1.3Meta | 175 tok/s | 1M | $1 | $4 | |
| 7 | Gemini 3.8 FlashGoogle | 125 tok/s | 1M | $0.50 | $2 | |
| 8 | Claude Opus 5.5Anthropic | 115 tok/s | 1M | $4 | $20 | |
| 9 | GPT-5.6 TerraOpenAI | 108 tok/s | 1M | $0.90 | $3.60 | |
| 10 | Claude Sonnet 5.5Anthropic | 104 tok/s | 1M | $3 | $15 | |
| 11 | Gemini 4 ArgonGoogle | 88.0 tok/s | 1M | $1.25 | $5 | |
| 12 | Step 5 Preview阶跃星辰 | 87.0 tok/s | 1M | $0.60 | $2.40 | |
| 13 | GLM-5.3智谱 | 83.0 tok/s | 1M | $0.80 | $3.20 | |
| 14 | Grok 4.7xAI | 74.0 tok/s | 500K | $3 | $12 | |
| 15 | Gemini 4 CarbonGoogle | 62.0 tok/s | 1M | $2.50 | $12 | |
| 16 | DeepSeek V4深度求索 | 60.0 tok/s | 128K | $0.28 | $1.10 | |
| 17 | GPT-6.1 SolOpenAI | 56.0 tok/s | 1M | $0.60 | $2.40 | |
| 18 | Qwen3.8-Flash-Next阿里巴巴 | 56.0 tok/s | 256K | $0.15 | $0.60 | |
| 19 | Claude Fable 5.2Anthropic | 54.0 tok/s | 1M | $5 | $25 | |
| 20 | Claude Fable 5.1Anthropic | 54.0 tok/s | 1M | $4 | $16 | |
| 21 | GLM-5.3-Flash智谱 | 53.0 tok/s | 1M | $0.25 | $1 | |
| 22 | GPT-6 AstraOpenAI | 43.0 tok/s | 1M | $2.50 | $10 | |
| 23 | MiMo-V2.6-Pro小米 | 43.0 tok/s | 1M | $0.30 | $1.20 | |
| 24 | Kimi K3月之暗面 | 41.0 tok/s | 1.05M | $0.70 | $2.80 | |
| 25 | Qwen3.8 2.4T A95B阿里巴巴 | 37.0 tok/s | 262K | $0.90 | $3.60 | |
| 26 | Qwen3.8 Max阿里巴巴 | 36.0 tok/s | 984K | $1.60 | $6.40 |
切换榜单视角
返回模型排行榜,查看全部榜单视角与收录说明。