LLM token speed simulator
Compare two simulated streams. Choose a speed and press Start.
Illustrative text chunks, not a model benchmark.Response length & presets
Choose your speeds, then start the comparison.
Changing a setting resets playback. Leaving this tab pauses the simulation.
First stream
ReadySecond stream
ReadyLane B finishes sooner.
How token timing works
Two speeds worth
telling apart.
Time to first token is the wait before the first piece appears. Tokens per second is the pace after that. This simulator lets you change both independently.
A shorter wait can matter most for a quick answer. A faster stream makes more difference as the response gets longer.
How is this calculated?
Total time = first-token delay + (tokens − 1) ÷ tokens per second.
The delay already includes the first token, so the remaining tokens arrive at evenly spaced intervals. Paused time does not count.
Are these real model tokens?
No. These are illustrative chunks of prepared English text, split into small pieces. A real tokenizer uses its own vocabulary and can produce different counts for different text and languages. The passage repeats for longer runs. This is a visual speed comparison, not a tokenizer or an estimate of API charges.
Does this predict my GPU or model speed?
No. It does not run a model, contact an AI service, or measure hardware. Real response times depend on the model, prompt, context length, hardware, load, network and the way output is delivered. A faster stream also does not tell you whether an answer is better.
Why do several pieces sometimes appear together?
Browsers redraw a screen in frames. At higher rates, several simulated tokens can arrive between frames. The counter follows elapsed simulation time rather than assuming one token per frame.