Tip: Interactive Token Speed Simulator Shows What 10-800 Tokens/Second Actually Feels Like

Simon Willison·low signalXBlueskyLinkedInCopy link

Mike Veerman built an interactive HTML tool (highlighted by Simon Willison) that simulates LLM token output at speeds from 5 to 800 tokens per second, helping developers viscerally understand the difference between model speed tiers. This matters for agent UX design — at 10 tok/s an agent feels sluggish, at 100 tok/s it feels responsive, and above 200 tok/s the bottleneck shifts to tool execution. Useful for calibrating model selection between Opus-class (slower, smarter) and Flash-class (faster, cheaper) for different agent phases.