What does 20 tokens per second feel like?
Watch text stream at any generation speed and compare it with how fast you read.
15words per second
This model
You, reading
~4 w/s
Why the number matters
Generation speed, in tokens per second, is the number every local AI benchmark leads with. It is hard to picture, so this tool plays text at that speed.
A token is a piece of a word. English averages about 0.75 words per token, so 20 tokens per second is about 15 words per second. Most people read around 4 words per second, which is why anything above roughly 15 tokens per second already feels fast for chat.
Speed is not the whole wait. Before the first word, the model has to read your prompt; the answer time tool shows that part.