Models
GPT-6 Astra, Claude Fable 5.1, Gemini 3.8 Flash, and leading open-weight models through one API. Choose up to five models for your agent, or use the default smart-routing lineup.
Missing a model? We continuously add new ones.
What the router weighs.
Use the default model set, or pin exactly the models you want from the dashboard. Within your set, the smart router figures out what performs best and sends each request there.
Among the models capable of the job, the router weighs current price and latency, and picks the one that gets you the most compute per dollar. It also biases toward recently used models and providers, so prompt caches stay warm and repeat work costs less.
The same model is often served by many providers at different speeds and prices. The router picks the best serving option per request and re-scores providers continuously, so you never juggle keys or manage fallbacks.
Every plan includes every model. Plans differ by the compute budget allocated, never by which models you can reach.