One model name. Every frontier lab.
You never pick a model. Set one name and the router matches each request to the right one — frontier when it counts, efficient when it doesn’t.*
model: "standardcompute"* A representative lineup, not the full roster. Models and providers shift as we review and re-score them — new releases often land before this page catches up.
What the router weighs.
Every request is read before it is routed. Routine traffic — classification, summaries, short answers — goes to efficient models; deep reasoning, hard code, and long-horizon work goes to the frontier.
Among the models capable of the job, the router weighs current price and latency. Not wasting frontier tokens on routine work is what keeps a flat rate sustainable.
Providers are monitored and re-scored continuously. When one degrades, eligible traffic shifts to an alternative — you never manage fallbacks or juggle keys.
Every plan includes every model — plans differ by speed, never by which models you can reach.