Tokens per second visualizer
See how fast an AI model writes at any speed
Runs in your browser. Nothing is uploaded.
How it works
Tokens per second measures how fast an AI model writes its answer. One English token is about three quarters of a word. At 5 tokens a second, text keeps pace with a fast reader. At 50, a long answer lands in seconds. Pick a speed and watch text stream at that rate.
words per second = tokens per second × 0.75
Language models write one token at a time. A token is a small piece of text, often part of a word, so speed is measured in tokens per second.
This tool plays sample text at the speed you choose, after the delay you set before the first token. It then shows how long the answer takes and how that compares with reading speed. We count 0.75 words to a token and take reading speed as 4 words a second.
Worked example. At 50 tokens a second, a 300-token answer takes 6.5 s, with a 500 ms wait for the first token.
Examples
| Tokens a second | Words a second | 300 tokens take | Feels |
|---|---|---|---|
| 5 | 3.75 | 60.5 s | About reading speed |
| 20 | 15 | 15.5 s | Faster than reading speed |
| 50 | 37.5 | 6.5 s | Faster than reading speed |
| 200 | 150 | 2 s | Faster than reading speed |
What this tool doesn’t cover
- Speeds here are what you enter, not a test of any real model.
- Word counts assume English. Other languages often use more tokens per word.
Questions
What is a good tokens per second speed?
For chat, around 20 to 50 tokens a second feels quick, because text appears faster than most people read. Long answers and code benefit from more.
How many words a second is 50 tokens a second?
About 37.5 words a second, because one English token is about 0.75 words.
What is time to first token?
The delay between sending a prompt and seeing the first word. It often affects how fast a chat feels more than the writing speed does.
Why does the text seem to arrive in bursts in some AI apps?
Many apps send tokens in small groups, and networks add delay. So the speed you see can differ from the model’s real rate.
Does a faster model mean better answers?
No. Speed and quality are separate. A smaller model is often faster, while a larger one may be slower but more careful.
Was this useful?
Built by QuickToolz.
Sponsored
Advertise here10 spots leftSponsored
Advertise here10 spots left