How DreamOS86 selects AI models, and how to customise model selection for your projects.
Auto mode
By default, DreamOS86 uses Auto mode to select the best model for each task. Auto routing considers:
Task type — simple edit vs. complex architecture
Context length — how many tokens your request requires
Budget — your remaining credits and plan tier
Speed requirement — whether you need a quick response or can wait for higher quality
Manual model selection
You can override the auto-selected model at any time using the model picker in the chat input bar.
Orchestration modes
When creating an app from the home screen, you can choose an orchestration mode:
| Mode | Runtime | Model strategy |
| ------ | --------- | ---------------- |
| Fast | 8–20s | Lightweight models throughout |
| Balanced | 25–45s | Mixes lightweight and standard |
| Deep Build | 50–90s | Standard + premium for architecture |
| Production | 80–120s | Premium models with quality verification |
| Autonomous | 90–180s | Agentic loop with self-review |
Model tiers
| ------ | -------- | ------------- |
| Standard | Claude Haiku, Gemini Flash, GPT-5.4 Mini | Low |
| Premium | Claude Sonnet, GPT-5.4, Gemini Pro | Medium |
| Ultra | Claude Opus, GPT-5.5, Grok 4 | High |
Your plan determines which tiers you can access:
Free & Starter: Standard only
Pro: Standard + Premium
Studio & above: All tiers including Ultra
Default model
You can set a default model in Settings → Models → Default Model. This model is used when Auto mode selects a standard-tier model, replacing it with your preference.
Disabling models
Disable specific models to prevent them from being used, even by Auto routing. Go to Settings → Models and toggle the models you don't want.
Recommended model per task
| Simple UI edits | Claude Haiku / Gemini Flash |
| Feature development | Claude Sonnet / GPT-5.4 |
| Full app generation | Claude Opus / GPT-5.5 |
| Large codebase analysis | Gemini Pro (2M context) |
| Code refactoring | Composer (Cursor) |