Which AI model writes your code — and can you choose?

Which model families can you choose between?

The coding tool lists eight model families at the point of choice: Grok, Claude, GPT, DeepSeek, Kimi, MiniMax, GLM and Qwen. None is presented as the only option, and the credit price is shown beside each name before you confirm.

Why this matters is practical. You may already hold a provider key for one of these families, or a team may prefer one engine's style for a particular codebase. The choice is visible before a build starts, not hidden in settings after the fact.

• Grok
• Claude
• GPT
• DeepSeek
• Kimi
• MiniMax
• GLM
• Qwen

How is the price shown before you commit?

A fair price display happens before the click, not on the bill. While you are choosing an engine, the interface shows the credit price of each option. You can compare the cost of running the same request on different engines without starting a build.

What changes when I bring my own provider key?

Bringing your own provider key changes which price applies. An engine that runs on your key runs at the base price. If the platform holds no key for an engine, that engine runs only on your own key. The key itself is not placed inside the build sandbox.

This matters because it moves the cost decision into the open. You can decide whether to use a key you already hold or let the platform handle the provider key for that engine.

Credit price

The amount shown for an engine before you choose; it is the cost when the platform supplies the key.

Base price

The price applied when an engine runs on your own provider key.

Provider key

An API key from a model provider that you supply.

Sandbox

An isolated, single-use environment with a shell, file search and real file edits; it is destroyed when the run ends.

When is a simpler tool the better choice?

If you only ever use one model family, never need a sandbox, and do not care about comparing credit prices before a run, a single-model coding assistant may be perfectly adequate. The benefit of model choice appears when you switch engines between tasks, use a provider key you already hold, or want the price visible before you commit.

If your work is a quick snippet rather than a real build, a lightweight tool might be faster to open. The case for this kind of coding surface is strongest when the model choice, price display and isolated build environment all change how you work.

What happens during a real build?

The coding tool can run a build in an isolated, single-use sandbox. The sandbox has a shell, file search and edits on real files, so the model can do more than return a snippet. The sandbox never holds a provider key. Model calls go through GROX, which holds the keys, and the sandbox is destroyed when the run ends. This keeps a real build contained without putting your keys inside the environment.