💤 Select a task and press Run Task
⏱ —
📖 How to read this simulation
- Each column = same model at a different generation speed
- 🤔 Thinking (yellow) = model processes the prompt, plans the answer — does NOT depend on tok/s, it's pure compute power
- ✍️ Generating (blue) = model writes the response — this IS where tok/s matters
- ✅ Done (green) = response complete
- Total time =
think_time + (output_tokens ÷ tok/s)
- For document analysis, think-time is large (10-30s) even if the response is short
- For simple chat, think-time is low but the response feels slow if tok/s is low
- "real: X tok/s" is measured from tokens actually rendered ÷ elapsed time
💤 Press ▶ Start All to begin
⏱ 0.0s
📖 Understanding token speeds
- 1 token ≈ 4 characters (~0.75 English words)
- 10 tok/s ≈ 40 chars/s ≈ 450 wpm — leisurely reading
- 16 tok/s ≈ 64 chars/s ≈ 720 wpm — fluid conversation
- 30 tok/s ≈ 120 chars/s ≈ 1350 wpm — fast reading
- 50+ tok/s ≈ 200+ chars/s — near-instant
- Real time is the same for all columns; the difference is how much text each one produced