Models
Pick a model per thread, set reasoning effort and speed, see what your plan includes, and connect ChatGPT or OpenCode.
Every chat thread runs on one model. The model picker sits at the left of the composer toolbar, and on the Home screen next to the prompt. It lists the models your plan and connected accounts allow; the list changes as models are added or retired, so treat the picker as the source of truth.
Choosing a model
- Click the model name in the composer toolbar (Select model).
- Hover or arrow through the list. A card beside the list shows the model's description, context size, prices per 1M tokens where they apply, the reasoning control and any reason it is unavailable.
- Click a model to use it.
The trigger shows the model name and, where set, its effort and speed: GPT-6-Sol · High · Fast (a ChatGPT model).
The list has up to three groups:
| Group | Billed to | Shows when |
|---|---|---|
| Embedr models | Your Embedr plan usage | Always |
| ChatGPT | Your ChatGPT subscription | You connected ChatGPT; otherwise a Connect ChatGPT row links to Settings |
| OpenCode Zen and OpenCode Go | Your OpenCode balance or Go subscription | You connected OpenCode; otherwise a Connect OpenCode row links to Settings |
The card's route label (Embedr, ChatGPT, OpenCode Zen, OpenCode Go) tells you which account a model draws from. The same model can appear in several groups.
Current headline models on the Embedr route include GPT-6 Sol, GPT-6 Luna and GPT-6 Astra, Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5, Kimi K3.1 Pro and DeepSeek V4.1 Flash. All of them accept images.
Per-thread model and the default for new threads
| Situation | Model used |
|---|---|
| You pick a model in a thread | That thread keeps it from now on, and it becomes the default for new threads |
| You start a new thread | It starts on your most recent pick |
| You switch back to an older thread | It keeps the model it was using, even if you picked another one elsewhere |
| A thread's model is retired or no longer available | Embedr switches the thread to an available model |
Picking a model on the Home screen sets the model for the project and thread it creates. Helpers use the thread's model unless their kind names another, and always a model your plan includes.
Reasoning effort
Models that support it show a Reasoning track in the hover card. Click a level:
| Level | Short label |
|---|---|
| None | None |
| Low | Low |
| Medium | Med |
| High | High |
| Extra high | X-High |
| Maximum | Max |
Not every model offers every level. Higher effort thinks longer before acting: slower and more usage per request, usually better on multi-step debugging, board work and unfamiliar code. Lower effort suits quick edits and questions.
The level is remembered per model on this device. Some models show Required (reasoning cannot be turned off). On the Free plan, GPT-6 Sol's effort is Fixed on Free.
Speed
ChatGPT models show a lightning button in the hover card. Fast responds faster and uses 2.5× ChatGPT credits; click again for Standard speed. The trigger shows · Fast while it is on.
What your plan includes
| Plan | Models |
|---|---|
| Free | GPT-6 Sol on the Embedr route, plus any ChatGPT or OpenCode models you connect |
| Hobby, Pro | The full Embedr list |
| Embedr Access | A smaller set of Embedr models, shown as available in the picker |
Models your plan does not include are greyed out with a lock and a badge (Hobby+ or Pro); the hover card says "View plans to use this model". If a request still reaches a model your plan does not include, the chat shows "Your plan doesn't include this model. Choose another model or upgrade your plan."
Embedr models count against your plan's usage. The usage meter in the composer toolbar (View usage) shows what remains and when it resets, with Upgrade plan or Manage billing. When usage runs out, the chat shows Model Usage Limit. See Settings and privacy for Usage & billing.
Some new models need a newer app version. If a model you expect is missing, update Embedr.
Connect ChatGPT
Use a ChatGPT Plus or Pro subscription for ChatGPT models.
- Open Settings → AI Accounts (or click Connect ChatGPT in the picker).
- Under OpenAI → ChatGPT Plus / Pro, click Connect.
- Click Copy code, then Open ChatGPT, and enter the one-time code in the ChatGPT window.
- Back in Embedr, the account shows as connected and the ChatGPT group appears in the picker.
Use the refresh button to re-check the connection and Sign out to disconnect. If ChatGPT rejects or expires the connection, the chat shows Reconnect ChatGPT; reconnect and retry.
Connect OpenCode
Use an OpenCode API key for OpenCode Zen and Go models.
- Open Settings → AI Accounts → OpenCode.
- Click Get Key if you need one, then paste it (
sk-...) and click Save Key. - The OpenCode Zen and OpenCode Go groups appear in the picker.
Zen and Go models draw from your OpenCode balance. Go models also need an active OpenCode Go subscription; without one they are greyed out with "An active OpenCode Go subscription is required." If your balance runs out, the chat shows OpenCode balance required. Embedr never picks an OpenCode model for you automatically; choose one explicitly.
Choosing well
- Attach screenshots, scope captures, board photos and schematics freely; every Embedr model reads images.
- Use a stronger model or higher effort for board bring-up, PCB and schematic work, and long debugging sessions.
- Use a fast model or low effort for small edits, explanations and quick lookups.
- When a long thread gets slow or nears its context limit (the ring in the composer), start a new thread and mention the files or spec you need.
Troubleshooting
| Message | Fix |
|---|---|
| "Log in to ChatGPT from Settings to use this subscription model" | Connect ChatGPT in Settings → AI Accounts |
| "Connect OpenCode from Settings to use this subscription model" | Save an OpenCode key |
| Provider Rate Limit | Wait a moment and retry, or switch models |
| "Model is overloaded or returned an invalid response. Please retry or pick another model." | Retry or switch models |