In short
- One setting.Switch an agent to Auto in its Model tab.
- Picks the model for you.Auto chooses the model, so you do not have to.
- Stronger for language.By default, language is the one thing that moves Auto up a tier.
- Keeps going if a model fails.If a model does not start answering, the next one takes over.
Not every agent needs a model picked by hand. Auto picks the model for you, so you do not have to trade quality against cost yourself.
How Auto decides
Models are grouped into tiers, from fast and low-cost to premium. Auto picks a tier by fixed rules, so the same inputs always lead to the same choice. By default, the only rule that moves it up a tier is language.
Within a conversation, Auto stays on the same model where it can, so the conversation flows and answers start faster. It only moves when a question needs a stronger model.
What Auto will not do
Auto chooses from the models OneType offers in the fast, standard and premium tiers. It does not use the largest, most expensive models. Pin one of those, or a Claude model, on the agent instead. Auto also never sends a conversation to a provider that needs your explicit opt-in.
It keeps answering when a model has a bad moment
AI providers sometimes slow down or fail. If a model does not start answering in time, OneType moves on to the next suitable model before your visitor sees an error. Your visitor just gets an answer.
How Auto is counted
In the model picker, Auto shows as about 1 credit per answer. Every answer is billed by the tokens it actually used, so a short answer on a light model can cost less and a harder one on a premium model can cost more. You can see where your credits go by day, agent and model in your analytics.
Languages and images
Language is the one thing that moves Auto up a tier by default: for a language where smaller models are less reliable, Auto uses a stronger model.
When a visitor attaches an image, Auto only chooses models that can read images. When your agent has actions it can take, Auto only chooses models that can use them. You never have to think about which model supports what.
Why a deterministic choice helps
Auto follows fixed rules, not a guess. The same kind of question in the same situation always goes to the same tier. That makes your agent predictable: if a type of question gets a strong answer today, it gets the same treatment tomorrow. And if you would rather have one model for everything, you can pin it on the agent instead.
What each plan includes, and how AI credits work, is on the pricing page.