The short answer
Auto reads your message, sorts it into one of six fixed rungs, and answers on the model pinned to that rung, never above what your plan can open. Auto carries no credit price of its own: a turn costs whatever the model Auto picked costs, which is why the clients publish a range for your tier instead of a single number. Those ranges are 1 to 1 credits on Starter, 1 to 10 on Pro, and 1 to 20 on Powerhouse.
Auto is a row in the model picker rather than a model, it is available on every paid plan, and it sits in the Recommended section beside the house model and one current GPT, Claude and Gemini row. That section is constrained to models reachable at Pro or below, so the top of the picker is never Powerhouse-locked.
The whole ladder, cheapest rung first. The model column is reference data, quoted as the catalogue writes it:
| Rung | What sends a message there | Model | Credits | Lowest plan that reaches it |
|---|---|---|---|---|
| Quick | Short questions and lookups | whizi-ai | 1 | Starter |
| Rewrite | Reformatting work such as rewrites and translation | google gemini-3.7-flash | 2 | Pro |
| Attachment | A turn carrying an image or an upload | google gemini-3.7-flash | 2 | Pro |
| Long-form | Large pasted text, over about 6,000 characters | openai gpt-5.6-terra | 4 | Pro |
| Code | Code | anthropic claude-sonnet-5 | 10 | Pro |
| Reasoning | Multi-step reasoning | openai gpt-5.6-sol | 20 | Powerhouse |
Six rungs across five model rows, out of a catalogue of 280+. Every other family in the catalogue is a hand-pick in the model picker.
What Auto can pick on each plan
Auto never sends a turn to a model your plan cannot open. It walks back down the ladder until it finds a row your subscription can reach, so a message you already sent is never interrupted by a paywall.
| Plan | Rungs it reaches | Published credit range |
|---|---|---|
| Starter, $15.99/month | Quick | 1 to 1 |
| Pro, $29.99/month | Quick through Code | 1 to 10 |
| Powerhouse, $49.99/month | All six | 1 to 20 |
Starter reaches exactly four picker entries: Auto, the house model Whizi AI, one base GPT row and one fast Gemini row. None of the five other rungs is in that set, so a Starter account stays on the house model at 1 credit no matter what it asks.
Pro tops out at 10 because the reasoning rung is a Powerhouse-only row. The fallback is deliberately not the credit order: a Pro subscriber who asks a reasoning question drops to the code model, which reasons well, rather than to the cheapest row on the ladder.
Auto is listed as available on every paid plan, and its lowest rung, the house model, is a Starter row. Free accounts are not model-gated on the chat path, so nothing blocks the Auto row there either: a free account can try any text model inside a lifetime allowance of 7 messages, which never resets. What a free account does not have is a credit balance. The credit limit on free is 0, so the 1 to 1 range published for free is what the unauthenticated picker endpoint ships, not a charge a free account ever incurs.
Routing to your own tier also removes one class of failure outright. Because Auto only ever picks from rows your plan can already open, a turn on Auto does not return "Upgrade your plan to use this model." That error is HTTP 403 with the code tier_upgrade_required, and it comes from choosing a model above your tier by hand. The refusals that do still apply on Auto are the ordinary allowance and burst refusals, and they read the same as on any model: see rate limits and model unavailable.
What a turn on Auto costs
Auto resolves to a concrete model before anything is charged, so the credit charge, the tier gate and the stored history all record the model that actually ran, not the word "auto".
The reason a range is published rather than a single figure is mechanical: the endpoint that serves the picker is unauthenticated and does not know who is asking, so it ships a per-tier map instead of one number.
A model's credit cost is a fixed integer per message. Nothing about message length, conversation length or answer length enters the calculation. Length changes what a turn costs only because it changes which rung Auto picks: a message over about 6,000 characters routes to the long-form rung at 4 credits, on the reasoning that a message that long is a pasted document rather than a question.
Because the amount varies per turn, Auto is the row where a whole-turn refusal is easiest to meet: a 20 credit reasoning turn needs 20 credits free, not whatever happens to be left. Running out mid work is covered in out of credits.
Credit metering is live on the web client. A platform that is not on the credit allowlist sees a flat 1x instead and is charged one per turn, so no Auto range applies there.
The full credit scale, with how many models sit on each rung, is in how credits work.
What is sticky inside a thread
Difficulty only moves up. The previous turn's model acts as a floor, so a follow-up to a reasoning answer cannot drop back to the cheapest model. A thread that has earned an expensive rung stays there for its follow-ups.
Attachments always route the same way. Any turn carrying an image or an upload goes to the Gemini Flash rung at 2 credits, on the reasoning that an image needs vision rather than a large context window. On Starter, which cannot reach that row, the same turn is answered on the house model at 1 credit.
Length is the one hard threshold. About 6,000 characters is where a message stops being treated as a question and starts being treated as a pasted document, which sends it to the long-form rung.
If you want a specific model rather than a rung, switching is a per-message decision. That is covered in switching models mid-conversation.
What Auto is not
Auto is not configurable. The ladder is six fixed rungs pinned to five model rows. Settings offers Subscription, Cancel, Reset password, Export data, Import chats, Default model, Dark mode, Language, and support and legal links, and nothing on that list edits the Auto ladder, weights a rung, or excludes a model from it.
Auto is not the same thing as the router models in the catalogue. The catalogue also carries provider router identifiers, and those are ordinary picker rows: they are charged a flat 20 credits by owner policy rather than by arithmetic, because a router picks a downstream model per request and has no fixed token price. A router that lands on the dearest Claude Opus row is still sold at 20. Those rows sit on Powerhouse, and Whizi Auto is not one of them.
Auto does not reach every family. Its five model rows are the house model, a Gemini Flash row, a mid GPT row, Claude Sonnet and a top GPT row. A DeepSeek, Grok, Qwen, Kimi, GLM, Llama or Mistral answer is always a hand-pick in the picker, at any tier.
The catalogue by plan is in the model list, and what each plan includes is in plans and limits.
- Auto has six fixed rungs across five model rows, and no credit price of its own
- A turn costs whatever the model Auto picked costs: 1 on Starter, 1 to 10 on Pro, 1 to 20 on Powerhouse
- Auto never picks a model above your plan, so it never returns "Upgrade your plan to use this model."
- A message over about 6,000 characters routes to the long-form rung at 4 credits
- Any turn carrying an attachment routes to the Gemini Flash rung at 2 credits
- Difficulty is sticky: a follow-up cannot drop back below the previous turn's model
- Because the cost varies per turn, a 20 credit turn needs 20 credits free before it will send
Frequently asked questions
How do I see which model answered on Auto?
Reload the chat. Auto resolves to a concrete model before the turn is charged, and the stored history keeps that model, so the turn shows the row that actually answered rather than the word "auto".
Is Auto cheaper than picking an expensive model myself?
Not at the top of the ladder. The reasoning rung and the top Claude Opus row both cost 20 credits, so choosing between them is a product call rather than a cost saving. Auto's ceiling is whatever rung your plan reaches: 10 credits on Pro, because the reasoning rung is Powerhouse-only, and 20 on Powerhouse.
Can I change what Auto picks, or pin it to one model?
No. The ladder is fixed at six rungs, and no setting edits it, weights a rung, or excludes a model from it. If you want a specific model, pick it in the model picker for that message.
Is Auto the same as the router models in the catalogue?
No. The catalogue's provider router identifiers are ordinary Powerhouse picker rows, charged a flat 20 credits by owner policy because a router picks a downstream model per request and has no fixed token price. Whizi Auto is not one of them: it has no rung of its own and charges you for the model it landed on.
Does Auto work on the free tier?
A free account is not model-gated on the chat path, so the picker is not what stops it. The allowance is: 7 messages for the life of the account, which never reset, and a credit limit of 0. Whether Auto or a model you picked yourself answered makes no difference to that count.