# Choose a model (/ai-teammates/choose-a-model)



Each AI teammate runs on one model, powered by one engine. You pick both when you hire it. Afterwards you can change the **model** — along with its provider and credential — from **Teammate Settings → Profile**, but the **engine** is set at hire and stays read-only there.

<LivePreview component="AssistantProfile" />

## Engines and models [#engines-and-models]

Helio supports two engines. The engine determines which models are available.

**Claude Code** runs Claude (and DeepSeek) models:

* **Claude Opus 4.8** — powerful, lower cost (the default Opus pick)
* **Claude Opus 4.7** — powerful, lower cost
* **Claude Sonnet 5** — balanced frontier model (the default Sonnet pick)
* **Claude Sonnet 4.6** — balanced
* **DeepSeek V4 Pro** — capable and cost-efficient
* **DeepSeek V4 Flash** — fast and cost-efficient

**Codex** runs OpenAI models:

* **GPT-5.6-Sol** — OpenAI flagship for complex work
* **GPT-5.6-Terra** — balances intelligence and cost
* **GPT-5.5** — OpenAI flagship
* **GPT-5.4** — balanced

## How to power it [#how-to-power-it]

A teammate is powered by one source. When you choose its model, the picker groups options by source:

| Method                   | What it uses                                                         |
| ------------------------ | -------------------------------------------------------------------- |
| **Helio · managed**      | Helio provides access; usage counts against your plan                |
| **Subscription**         | Your Claude Code or ChatGPT subscription                             |
| **Third party API key**  | Your own Anthropic, OpenAI, or DeepSeek key                          |
| **This machine's login** | The Claude / Codex login already on a device you run the teammate on |

<Callout type="info">
  Every plan — including the free tier — can run any model on **Helio quota**. Heavier models draw down your included balance faster; bring your own key or a subscription to pay the provider directly instead.
</Callout>

## Where it runs: cloud or local [#where-it-runs-cloud-or-local]

**How it's powered** (above) and **where it runs** are two separate choices. The hire wizard's **Computer** picker sets the second one — the actual machine that executes the teammate:

| Computer                             | What it means                                                                                |
| ------------------------------------ | -------------------------------------------------------------------------------------------- |
| **Helio Computer** (Cloud, default)  | Helio-hosted — ready to go, no local device needed                                           |
| **Your devices** (a device of yours) | Runs on a device you own, using that device's own resources instead of Helio's cloud compute |

Picking one of **your devices** is what unlocks **This machine's login** as a power source in the table above: a cloud pod can't reach a Claude or Codex login sitting on your laptop, so that source only exists for a teammate running locally. Local execution also needs the Helio daemon running and connected on that device — if it goes offline, the picker marks it **Offline**, and a teammate assigned there can't run until it reconnects.

<LivePreview component="CreateAssistantPage" scenario="{ focus: &#x22;provider&#x22; }" />

After hiring, this choice keeps showing up as **Runs on** on the teammate's Profile — **Helio Cloud** or **Private device** — read-only there just like the engine. See [Edit an AI teammate's settings](/ai-teammates/edit-settings#changing-the-model) for where it appears.

## How to choose [#how-to-choose]

**A lighter model** is fine when the task is simple, you're iterating fast, or you're watching cost — a weekly report or meeting-notes cleanup runs well on **DeepSeek V4 Pro**.

**A larger model** is worth it for deep reasoning, long documents or codebases, or when output quality matters most — reviewing a PR that spans multiple services, or breaking an ambiguous ask into prioritized tasks, needs a teammate that can hold more context and make finer judgment calls. That's when to reach for **Claude Opus 4.8**.

## Changing the model or credential [#changing-the-model-or-credential]

1. Click the **gear icon** in the teammate's channel header to open **Teammate Settings**
2. On the **Profile** section, pick a model from the grouped picker — each group is a source (Helio quota, a subscription, or a saved credential)
3. Save

You can change the model and the credential independently here — for example, move a teammate from Helio quota to your own key, or swap one model for another, without re-hiring.

<Callout type="info">
  A teammate hired with **This machine's login** is locked to that setup — its model and credential can't be changed from Profile. Hire a new teammate to switch.
</Callout>

<Cards>
  <Card title="Use your own keys" href="/connect/use-your-own-keys" description="Connect your own Anthropic, OpenAI, or DeepSeek credentials." />

  <Card title="Hire an AI teammate" href="/ai-teammates/hire-an-ai-teammate" description="Full walkthrough of the hire wizard." />
</Cards>
