Models and Auto

The right modelfor every turn

Run on the subscriptions you already pay for, or on AQ models with the price shown before you pick. Auto routes each turn across both and shows its choice, price and confidence in the transcript.

Auto's routing decisions in an AQ transcript: the model chosen for each turn with its price, confidence and task type

Every AQ thread runs on a model you choose. Pick one of your own subscriptions (Claude Code, Codex, Cursor, Kimi Code, Grok, Antigravity, Devin, Muse Code, OpenCode, Command Code) and it runs on that plan with no AQ meter. Pick an AQ model and it is billed in AQ credits at the per-token price shown in the menu. Or pick Auto and let AQ choose per turn.

Three ways to pick

Your subscription

Unmetered

Sign in to your own accounts once. AQ never marks up usage on them, and you can connect several accounts of one service.

An AQ model

Priced up front

Every named model shows its own per-token price in the composer before you pick it. Pro includes $10 of AQ credits per seat each month.

Auto

Routed per turn

One decision per turn across AQ models and your subscriptions that have room left, written on the transcript.

A transcript row showing Auto's choice of model, its price per million tokens, its confidence and the task type
Every Auto turn says which model it chose, what that costs, how sure it was and what kind of task it read.

How Auto chooses

Before each turn, Auto reads the conversation and weighs every model it may use: AQ models on lanes your organization allows that fit the context, and your connected subscriptions, which count as free only while the plan shows included usage left. It records the decision with its reason and confidence. When it is not sure, it asks you which way to go instead of guessing. If a turn fails, it can try once more on another model. Background: model routing for coding agents.

Effort and speed

Set effort per message. Where a model and your account offer it, a Fast or Ultrafast speed shows up beside effort, with a note on what it costs in plan usage or credits.

A model for security work

Hinoki is an offensive-security model on AQ. Use it to scan a diff, a branch or a whole codebase for exploitable vulnerabilities, as a crew's security reviewer, or as a weekly bot. It asks you once to confirm you are authorized to test the systems you point it at.

Switch any time

Change the model, effort or account in the middle of a thread. The conversation carries over, so you can start on one model and finish on another.

Bring your own subscriptions, or start on AQ models.

Start freeSee pricing

Frequently asked questions

Does AQ charge for my own Claude or ChatGPT subscription?

No. Threads on your own subscriptions run on your plan, and AQ never marks up that usage. AQ credits apply only to AQ models.

What does Auto do?

Auto chooses a model for each turn from the AQ models your organization allows and the subscriptions you connected that have room, and writes its choice, price and confidence on the transcript.

How much do AQ models cost?

Each named model shows its per-token price in the composer before you pick it. Pro includes $10 of AQ credits per seat every month; after that, usage is billed at those prices.

Can I change models mid-conversation?

Yes. Switch the model, effort, speed or account in the same thread and the conversation continues.