LoopCodeGet started
Models and effort
The hosted models LoopCode offers on each plan, how to choose one, and how reasoning effort trades speed for depth.
Hosted models
LoopCode hosts the models and includes them in your plan. You never bring an API key.
| Model | Maker | Context window | Plan |
|---|---|---|---|
| Auto | Picks a model per request | — | Free |
| Opus 5 | Anthropic | 1M tokens | Free and up |
| DeepSeek V4.1 Flash | DeepSeek | 1M tokens | Free and up |
| GPT-6 Astra | OpenAI | 1M tokens | Pro and Max |
| GPT-6.1 Sol | OpenAI | 1.05M tokens | Pro and Max |
| Grok 4.7 | xAI | 500K tokens | Pro and Max |
The model picker always shows the current list for your plan, with each model's maker.
Auto
On the Free plan, Auto is the default. It picks a model for each request so that quick questions go to a fast model and demanding work goes to a stronger one, which makes the Free allowance go further. Pro and Max have no Auto: every request goes to the model you chose.
Choose a model
| Where | How |
|---|---|
| Terminal | /model in a conversation, or loopcode -m <id> when you start |
| Desktop app | The model picker in the composer; Settings → Models sets the default for new sessions |
| VS Code | The model picker in the LoopCode panel |
loopcode -m gpt-6-astra
loopcode -p "Summarise the open TODOs" -m deepseek-v4.1-flash
A conversation keeps its model until you change it.
Reasoning effort
Effort sets how hard the model thinks before it acts: low effort answers quickly, high effort plans more and checks more.
| Level | Use it for |
|---|---|
low | Quick questions, small edits |
medium | Everyday work (the default) |
high | Tricky bugs, multi-file features |
extra | Hard problems where a wrong turn is costly |
max | The deepest reasoning the model offers |
Set it with /effort or --effort <level> in the terminal, alt , and alt . to step it down and up, or Settings → Agent behavior in the desktop app.
Adaptive effort
With adaptive effort on, effort follows the work: LoopCode goes a level up to plan, to check its work, and when it is stuck, and back down for routine steps. Turn it on or off with /effort adaptive on or /effort adaptive off.
Seeing the model's reasoning
/thinking shows or hides the model's reasoning in the terminal. In the desktop app, the same choice is under Settings → Appearance.