Tokens per second
Tokens per second (TPS) is the speed at which a model generates its answer, measured as output tokens per second for a single request. Together with time to first token, it decides how fast a response feels.
Why it matters for companies
The same model can run at very different speeds on different providers and hardware, so speed is a property of the provider as much as of the model.
For agents that run many steps in a row, slow output adds up to minutes of waiting per task, multiplied across every person on the team.
Example
An agent runs 40 steps in a row and each step writes about 500 tokens. At 50 tokens per second the generation alone takes almost 7 minutes. At 150 tokens per second it takes a little over 2 minutes.
How Standard Compute handles it
A single API gives you frontier and open-source models, and it routes each step to the cheapest model that can do it and the cheapest provider fast enough to serve it.