The right model for every task, and the receipt.
Auto starts every task on the cheapest model that can do it, and moves up a tier only when your checks still fail after a fix, Dus gets stuck or the model refuses. Or pin a model, and its effort, from the top bar. Project details shows what each session cost and saved.
- Tiers
- fast · standard · frontier
- Escalation
- On failed checks, repeated failure or refusal; never down
- Savings
- Measured per task against the top model
- Providers
- Anthropic, OpenAI, Gemini, DeepSeek, local, any OpenAI-compatible



