You choose which model powers a task, and the right choice trades off capability, speed, and cost. This article is about how to think about picking a model — not which exact one to use, since the roster changes often.
Pick by strength, not by name
The available models change over time, so match a model to the job rather than memorizing names:
- Hard work — complex code, planning, tricky debugging — is worth your most capable model, even if it's slower or costs more.
- Routine and background work — summaries, short edits, and the app's behind-the-scenes helper tasks — can use a faster, cheaper model without losing much.
Effort level
Many models let you set an effort (or thinking) level for a turn. Turn it up when a problem is genuinely hard and you want the model to work through it; leave it lower for simple tasks to save time and cost.
Providers, including a local option
The roster spans providers — Claude, Codex (GPT), and DeepSeek — and you can also run local models through Ollama for offline or privacy-sensitive work. Some models may be gated depending on your access.
Assign models per task, once
You don't have to choose per message. Under Settings → Chat you assign a model to each kind of task, so writing code gets a capable model while background helpers get cheaper ones. See Settings you should review once.
Use both agents
Claude and Codex have different strengths, so using both catches more than either alone — and lets you keep working by switching to the other when you hit a usage limit on one.
Next steps
- Settings you should review once — assign models per task.
- Reviewing code with multi-model pipelines — put more than one model on a review.
- Core concepts & vocabulary — the terms behind models and effort levels.