AI 模型速度排行榜
按实测输出吞吐(tokens/秒)与平均延迟排名。
- OpenAI o3-mini — 吞吐: 3022.0 tps, 延迟: 2562.33s
- OpenAI GPT-5 — 吞吐: 2399.0 tps, 延迟: 27005.84s
- ByteDance Doubao-Seedream-4.5 — 吞吐: 1072.4 tps, 延迟: 13428.00s
- ByteDance Seedance 2.5 — 吞吐: 1014.2 tps, 延迟: 362962.00s
- ByteDance Doubao-Seedream-5.0-lite — 吞吐: 761.3 tps, 延迟: 18915.00s
- ByteDance Seedance 2.0 — 吞吐: 528.0 tps, 延迟: 206241.50s
- ByteDance Seedance 2.0 Fast — 吞吐: 442.9 tps, 延迟: 91652.00s
- Google Gemini 3.5 Flash — 吞吐: 291.7 tps, 延迟: 17691.98s
- ByteDance Doubao-Seedream-5.0-pro — 吞吐: 224.6 tps, 延迟: 23223.67s
- OpenAI chatgpt-image-latest — 吞吐: 202.1 tps, 延迟: 21885.00s
- OpenAI GPT Image 2.5 Dev — 吞吐: 188.6 tps, 延迟: 33132.37s
- OpenAI gpt-5-nano — 吞吐: 188.6 tps, 延迟: 6508.58s
- Google Nano Banana (Gemini 2.5 Flash Image 🍌) — 吞吐: 184.3 tps, 延迟: 6397.25s
- Google Gemini 3 Flash Preview — 吞吐: 166.1 tps, 延迟: 6361.44s
- Google Gemini 3.6 Flash — 吞吐: 162.5 tps, 延迟: 10279.66s
- OpenAI dall-e-3 — 吞吐: 153.2 tps, 延迟: 27251.00s
- Google Gemini 3.1 Flash-Lite — 吞吐: 128.3 tps, 延迟: 4417.59s
- xAI grok-3 — 吞吐: 127.0 tps, 延迟: 457.00s
- xAI Grok 4.20 Multi-Agent — 吞吐: 121.3 tps, 延迟: 0.00s
- OpenAI gpt-4-gizmo-* — 吞吐: 120.7 tps, 延迟: 3515.28s
- Google Gemini 3.1 Pro Preview — 吞吐: 119.5 tps, 延迟: 20090.35s
- Z.ai GLM 5.3 Flash — 吞吐: 115.6 tps, 延迟: 3074.07s
- DeepSeek DeepSeek V4.1 Flash — 吞吐: 115.5 tps, 延迟: 6233.81s
- ByteDance Seedance 2.0 Mini — 吞吐: 114.6 tps, 延迟: 434151.00s
- xAI grok-4-1-fast-reasoning — 吞吐: 106.8 tps, 延迟: 0.00s
- xAI Grok 4.3 — 吞吐: 105.3 tps, 延迟: 1188.67s
- OpenAI GPT-5.2 — 吞吐: 103.8 tps, 延迟: 4431.78s
- OpenAI GPT Image 1.5 — 吞吐: 102.8 tps, 延迟: 65911.50s
- OpenAI GPT Image 2 Develop — 吞吐: 102.3 tps, 延迟: 29715.50s
- Alibaba qwen3-30b-a3b-thinking-2507 — 吞吐: 100.9 tps, 延迟: 0.00s
- DeepSeek DeepSeek-V4-Flash — 吞吐: 92.1 tps, 延迟: 2831.14s
- Google Gemini 3 Pro Preview — 吞吐: 91.2 tps, 延迟: 0.00s
- Alibaba qwen3-30b-a3b — 吞吐: 91.0 tps, 延迟: 0.00s
- OpenAI GPT-5.6 Luna — 吞吐: 89.9 tps, 延迟: 2904.00s
- Alibaba qwen3-next-80b-a3b-thinking — 吞吐: 89.5 tps, 延迟: 0.00s
- Google Gemini 3.8 Flash — 吞吐: 89.3 tps, 延迟: 10923.26s
- xAI grok-4-0709 — 吞吐: 88.7 tps, 延迟: 963.33s
- xAI grok-4-fast-reasoning — 吞吐: 88.0 tps, 延迟: 0.00s
- Anthropic Claude Claude Opus 5 — 吞吐: 86.8 tps, 延迟: 23821.54s
- xAI grok-3-mini — 吞吐: 86.4 tps, 延迟: 0.00s
- OpenAI GPT Image 1.0 — 吞吐: 83.7 tps, 延迟: 52152.00s
- xAI Grok 4.20 (Non-Reasoning) — 吞吐: 83.0 tps, 延迟: 814.00s
- Anthropic Claude Claude Fable 5.1 — 吞吐: 82.5 tps, 延迟: 40556.33s
- OpenAI GPT-5.6 Terra — 吞吐: 79.5 tps, 延迟: 4957.86s
- Alibaba qwen3-vl-plus — 吞吐: 77.2 tps, 延迟: 523.00s
- OpenAI GPT-5.5 — 吞吐: 76.9 tps, 延迟: 11122.78s
- xAI Grok Build 0.1 — 吞吐: 76.3 tps, 延迟: 0.00s
- Google gemini-2.5-flash — 吞吐: 75.6 tps, 延迟: 1660.00s
- Google Nano Banana 2 (Gemini 3.1 Flash Image 🍌) — 吞吐: 74.4 tps, 延迟: 14640.50s
- Google Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image 🍌) — 吞吐: 74.4 tps, 延迟: 20080.25s
- xAI Grok 4.20 — 吞吐: 74.2 tps, 延迟: 381.00s
- DeepSeek DeepSeek-V4-Flash 0731 — 吞吐: 73.1 tps, 延迟: 1736.33s
- xAI grok-4-1-fast-non-reasoning — 吞吐: 70.9 tps, 延迟: 552.06s
- Google Gemini 3.7 Flash — 吞吐: 68.6 tps, 延迟: 6609.20s
- Alibaba qwen3-235b-a22b-thinking-2507 — 吞吐: 64.0 tps, 延迟: 0.00s
- OpenAI GPT-5.1 — 吞吐: 62.5 tps, 延迟: 8888.38s
- Anthropic Claude Claude Sonnet 5 — 吞吐: 61.8 tps, 延迟: 5445.27s
- Z.ai GLM 4.7 — 吞吐: 61.6 tps, 延迟: 403.00s
- Z.ai GLM 5.2 — 吞吐: 60.7 tps, 延迟: 4713.23s
- Z.ai GLM 5.3 — 吞吐: 60.5 tps, 延迟: 4125.18s
- OpenAI gpt-image-1-mini — 吞吐: 60.5 tps, 延迟: 17468.18s
- OpenAI GPT-5.6 Sol — 吞吐: 60.4 tps, 延迟: 4967.50s
- Alibaba qwen3-8b — 吞吐: 60.4 tps, 延迟: 0.00s
- Anthropic Claude Claude Fable 5 — 吞吐: 59.7 tps, 延迟: 4435.09s
- xAI Grok 4.5 — 吞吐: 56.6 tps, 延迟: 4499.53s
- Z.ai GLM 5.1 — 吞吐: 56.0 tps, 延迟: 1271.33s
- MiniMax minimax-m2 — 吞吐: 55.3 tps, 延迟: 0.00s
- Z.ai GLM 5 — 吞吐: 55.0 tps, 延迟: 1403.00s
- OpenAI GPT-5.4 Mini — 吞吐: 53.3 tps, 延迟: 1945.55s
- DeepSeek deepseek-v3.1 — 吞吐: 53.0 tps, 延迟: 22527.00s
- Google Gemini 3.1 Pro Preview Customtools — 吞吐: 49.9 tps, 延迟: 8804.00s
- OpenAI GPT-5.4 — 吞吐: 48.9 tps, 延迟: 11061.70s
- OpenAI gpt-4o-2024-08-06 — 吞吐: 48.4 tps, 延迟: 0.00s
- Alibaba qwen3-235b-a22b — 吞吐: 48.0 tps, 延迟: 0.00s
- OpenAI GPT-6 Astra — 吞吐: 47.2 tps, 延迟: 3460.28s
- xAI Grok 4.6 — 吞吐: 45.8 tps, 延迟: 9564.93s
- DeepSeek DeepSeek-V3.2 — 吞吐: 45.8 tps, 延迟: 1384.75s
- OpenAI gpt-4o — 吞吐: 45.0 tps, 延迟: 4407.89s
- Anthropic Claude Claude Opus 4.8 — 吞吐: 44.3 tps, 延迟: 5175.55s
- OpenAI GPT Image 2.5 Sunburst — 吞吐: 44.1 tps, 延迟: 34579.56s
- OpenAI GPT Image 2.5 Flare — 吞吐: 43.8 tps, 延迟: 19109.50s
- DeepSeek DeepSeek-V4-Pro-0813 — 吞吐: 41.6 tps, 延迟: 1634.63s
- Moonshot kimi-k2-0905-preview — 吞吐: 41.5 tps, 延迟: 1020.75s
- OpenAI GPT Image 2 — 吞吐: 41.2 tps, 延迟: 33881.76s
- OpenAI gpt-3.5-turbo-16k — 吞吐: 40.7 tps, 延迟: 0.00s
- Moonshot kimi-k2-0711-preview — 吞吐: 40.6 tps, 延迟: 989.50s
- Google gemini-2.5-pro — 吞吐: 40.3 tps, 延迟: 21041.00s
- Alibaba Qwen3.6 Plus — 吞吐: 39.5 tps, 延迟: 0.00s
- Moonshot kimi-k2-thinking — 吞吐: 39.4 tps, 延迟: 1028.50s
- Anthropic Claude Claude Opus 4.7 — 吞吐: 38.8 tps, 延迟: 4561.98s
- ByteDance Doubao-Seed-2.1-pro — 吞吐: 38.0 tps, 延迟: 0.00s
- Moonshot Kimi K2.5 — 吞吐: 37.2 tps, 延迟: 1708.00s
- Moonshot kimi-k2-turbo-preview — 吞吐: 36.7 tps, 延迟: 991.00s
- OpenAI gpt-4.1-nano — 吞吐: 36.6 tps, 延迟: 1572.50s
- Alibaba Qwen3.5 Plus 2026-02-15 — 吞吐: 36.6 tps, 延迟: 0.00s
- Alibaba qwen3-32b — 吞吐: 35.5 tps, 延迟: 0.00s
- Moonshot kimi-k2-thinking-turbo — 吞吐: 35.3 tps, 延迟: 4550.50s
- DeepSeek deepseek-v3 — 吞吐: 34.5 tps, 延迟: 3081.57s
- Anthropic Claude Claude Sonnet 4.6 — 吞吐: 33.8 tps, 延迟: 7809.40s
- OpenAI gpt-5-mini — 吞吐: 33.5 tps, 延迟: 2902.94s