Skip to main content

Choosing an AI model

Studio lets you pick which AI handles your requests. There are three families to choose from, each with a few options at different capability and cost levels. The default works well for most work — you only need to swap if you have a reason to.

Where to pick​

Look near the message box at the bottom of the chat panel. There's a small button showing the name of the current model. Selecting it opens a Choose Model dialog where you can pick a different one. The choice applies from your next message onward.

To change the default that new chats start with, open Settings and look for the AI Model section.

What's available​

Claude (Anthropic)​

  • Opus — Most capable. Best at long, complex changes and tricky reasoning. The most expensive option.
  • Sonnet — Balanced. The default. A very good all-rounder and what most users should stick with.
  • Haiku — Fast & affordable. Quick and cheap. Great for small edits, copy tweaks, and simple changes.

GPT (OpenAI)​

  • GPT-5.4 — Most capable. OpenAI's top model.
  • GPT-5.4 Mini — Balanced.
  • GPT-5.4 Nano — Fast & affordable.

DeepSeek​

  • DeepSeek V4 Pro — Lowest cost. A capable model at the cheapest per-token rate available in Studio.

Which should I use?​

  • Most users, most of the time: stick with the default (Sonnet). It's a good balance of capability and cost and rarely lets you down.
  • Quick, simple edits where you want to save money: Haiku, GPT-5.4 Nano, or DeepSeek V4 Pro.
  • A large refactor, a tricky design problem, or something the default has been struggling with: Opus or GPT-5.4.
  • You prefer one provider over another: any of them are fine — they all do the same kinds of work.

It's fine to switch models within a single project, even within a single conversation. The model you pick handles your next message onward; previous messages stay as they were.

What about cost?​

Different models cost different amounts per token, and the gap between the cheapest and the most expensive is large — choosing carefully can meaningfully cut your bill on the Managed plan. As a rule of thumb, the "Most capable" models cost the most and the "Fast & affordable" / "Lowest cost" models cost the least.

How AI usage is billed has the details on tokens, your balance, and topping up.

Plan & model availability​

  • Managed plan — all models are available, billed per token at platform rates.
  • BYOK plan — Claude and GPT models run on the API keys you provide. DeepSeek isn't available on BYOK.

See Plans for a side-by-side comparison of the two plans.

Next​