Speed

How fast output arrives — sustained throughput and time to the first token, which trade off against each other and against cost. Measured on a provider's serving of the model, so it describes an endpoint rather than the weights.

Ordered by how many of these benchmarks place a model in their top 10, then by its best placing. Each column keeps its own order; nothing here is averaged.

Model	In top 10	Artificial Analysis LLM Leaderboard Output Speed	Artificial Analysis LLM Leaderboard Time To First Token
Celeris-1 celeris	2/2	#1 1573/s	#3 0.58s
Command A+ cohere	1/2	#19 196/s	#1 0.39s
Mercury 2 inception	1/2	#2 924/s	#94 3.54s
North Mini Code cohere	1/2	#105 27/s	#2 0.53s
Ling 3.0 Flash inclusionai	1/2	#3 407/s	#85 2.70s
Gemini 3.5 Flash-Lite google	1/2	#4 407/s	#102 8.96s
Ministral 3 3B mistral	1/2	#16 204/s	#4 0.63s
Gemini 3.7 Flash google	1/2	#5 394/s	#27 1.04s
Gemma 4 E4B google	1/2	#86 58/s	#5 0.65s
LFM2.5-VL-1.6B liquidai	1/2	#6 346/s	#33 1.13s
Ministral 3 8B mistral	1/2	#52 114/s	#6 0.74s
LFM2.5-8B-A1B liquidai	1/2	#7 339/s	#44 1.46s
Qwen3.5 9B alibaba	1/2	#66 86/s	#7 0.77s
Nemotron 3 Nano Omni 30B A3B Reasoning nvidia	1/2	#8 326/s	#25 1.00s
Granite 4.1 8B ibm	1/2	#53 112/s	#8 0.77s
Nemotron 3.5 Lightning nvidia	1/2	#9 314/s	#24 0.96s
Llama 4 Scout meta	1/2	#41 133/s	#9 0.79s
Nova Micro aws	1/2	#10 273/s	#21 0.90s
Muse Glimmer meta	1/2	#59 104/s	#10 0.79s

A dash means the source does not cover that model, not that it scored zero. Models with no registry entry are excluded here because they cannot be joined across sources — they are still shown, flagged, on their own source's board.