Different work wants different models. You can now choose — and pay the way that makes sense for you.
What changed
- A fast default, included. Every plan ships with a capable default model and a generous monthly allowance, so you can start building right away at no extra cost.
- Bring your own key. Add your own API key to run frontier models — Claude, GPT, and others — on your tasks.
- Switch per task. Pick the model in the composer for the job in front of you.
- Billed at provider cost. When you use bundled inference, it's passed through at the provider's rate — never marked up — and you set a per-task spend limit so a run can't drift past what you intended.
Why
We'd rather earn your trust on the platform than on a model markup. Use the best model for the moment, keep the bill predictable, and finish at a reviewed PR you merge.