Where you pick a model
- In Code, the model menu sits under the chat box, next to the mode and Effort. Each chat can use its own model. Settings, AI, Default Model sets the one new chats start with; out of the box it's Claude Sonnet 5.
- In Visual, you don't need to pick: Fast and Auto choose for you. Customize lets you choose the model yourself. See New project options.
- Settings, AI, Catalog lists every model, recommended ones first, with what each one is for and what it uses. You can set the default from a row.
Every listed model can work as an agent: it reads, writes and uses tools. Each menu entry says what the model is for and what it uses, so you can choose by job and by cost.
Recommended models
| Model | Maker | Context | Good for |
|---|---|---|---|
| Claude Opus 5.5 | Anthropic | 1M | The deepest reasoning: hard changes, architecture |
| Claude Sonnet 5.5 | Anthropic | 1M | A balance of speed and depth, everyday work |
| Claude Haiku 5.5 | Anthropic | 1M | The fastest Claude, quick questions and small edits |
| GPT-6 Astra | OpenAI | 1M | The hardest work |
| GPT-5.6 Sol | OpenAI | 1M | OpenAI's flagship |
| Gemini 4 Argon | 1M | The hardest problems | |
| Gemini 3.1 Pro | 1M | Agentic coding | |
| Gemini 3.8 Flash | 1M | Fast everyday work | |
| Grok 4.7 | xAI | 500K | Coding and agents |
| Mistral Large 4 | Mistral | 1M | Flagship generalist, European |
| Mistral Large 3 | Mistral | 256K | General purpose, European |
| Devstral 2 | Mistral | 256K | A coding agent, European |
"Context" is how much the model can read at once, in tokens (1M is about a million).
The full catalog
Beyond the recommended ones, the catalog has more from each maker, under Show more in the menus:
- Anthropic: Claude Fable 5.1 and Fable 5 (long tasks with many steps), Opus 5, Sonnet 5, Opus 4.8, 4.7, 4.6 and 4.5, Sonnet 4.6 and 4.5, Haiku 4.5.
- OpenAI: GPT-6 Luna, GPT-5.6 Terra and Luna, GPT-5.5, GPT-5.4 (with Mini and Nano), GPT-5.2, GPT-5.1, the GPT-5 family, GPT-4.1, GPT-4o, o3 and o3 Pro.
- Google: Gemini 3.7, 3.6 and 3.5 Flash, Flash-Lite models, Gemini 3 Flash.
- xAI: Grok 4.6, 4.5, 4.3, Grok 4.20 (reasoning and fast), Grok Build 0.1.
When a maker stops running a model, it leaves Piye's lists on that day by itself.
Models for approved accounts
Claude Mythos 5.1 and GPT-5.6 Cyber are offered by their makers only to approved organizations. They aren't available on Piye's credits. If your organization is approved, add its API key and they run on that key.
Models on Starter
On Starter, Code runs these models on Piye's credits: Claude Opus 4.6 and 4.5, Claude Sonnet 4.6 and 4.5, Claude Haiku 4.5, Devstral 2 and Mistral Large 3. The default on Starter is Claude Sonnet 4.6. Every other model shows in the menu with what opens it: a paid plan, or your own API key for that maker. Visual builds on Fast.
Local models
With Ollama on your computer, Code can use any model you have downloaded there. They show as "On this computer". See Your own API keys and local models.
Choosing well
- Start with the default. Move up to a bigger model when a task is hard, and down to a faster one for small edits.
- The Effort control in Code (for models that reason) trades depth for speed and credits. See Models, effort and keys.
- A long chat costs more per reply than a new one. Start a new chat for a new task.