速度

輸出有多快：持續輸出速度和首個 token 的等待時間，兩者之間、以及與價格之間都需要權衡。測的是某個服務供應商提供的 API，而不是模型本身。

模型	進入前 10	Artificial Analysis LLM Leaderboard Output Speed	Artificial Analysis LLM Leaderboard Time To First Token
Celeris 1 Celeris	2/2	#1 1523/s	#6 0.56s
Nemotron 3.5 Lightning NVIDIA	2/2	#10 307/s	#7 0.57s
Gemini 2.5 Flash Lite Google	1/2	#12 289/s	#1 0.32s
Mercury 2.5 Inception	1/2	#2 818/s	#135 3.21s
Gemini 2.5 Flash Google	1/2	#28 208/s	#2 0.46s
Mercury 2 Inception	1/2	#3 456/s	#148 6.51s
Claude Haiku 4.5 Anthropic	1/2	#95 90/s	#3 0.47s
Gemini 3.5 Flash Lite Google	1/2	#4 365/s	#151 8.96s
Command A+ Cohere	1/2	#35 175/s	#4 0.48s
Trinity Large Thinking Arcee AI	1/2	#5 339/s	#60 1.21s
Granite 4.2 IBM	1/2	#25 215/s	#5 0.53s
Ling 3.0 Flash InclusionAI	1/2	#6 334/s	#116 2.50s
Muse Spark 1.2 Meta	1/2	#7 325/s	#154 12.05s
Gemini 3.1 Flash Lite Google	1/2	#8 323/s	#146 5.54s
Ministral 3 Mistral	1/2	#23 222/s	#8 0.59s
Ling 3.0 Flash Fin InclusionAI	1/2	#9 321/s	#80 1.70s
Grok 4.20 SpaceXAI	1/2	#84 103/s	#9 0.59s
North Mini Code Cohere	1/2	#145 49/s	#10 0.60s