tokenspeed is a tool that lets you visually experience LLM token generation at different speeds, from 5 tok/s (Raspberry Pi-class) to 800 tok/s (Cerebras-class). It offers four modes — code, text, think, and agent — to simulate real streaming scenarios. The tool highlights how the same tokens-per-second rate feels perceptually different depending on content type, since code is more token-dense than prose.
134 Impressions