> For the complete documentation index, see [llms.txt](https://docs.instruqt.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.instruqt.com/ai-capabilities/connect-ai-models.md).

# Connect AI Models

Use AI models in your content via your cloud accounts

You can connect your own AI models to the platform and use them directly inside your labs and training content. This means you can build AI-focused content and hands-on exercises using the same models your organisation already runs in production.

To connect a model, you bring it in through a supported cloud account. Currently supported providers are Google Vertex AI and Amazon Bedrock.

***

#### How to Connect a Model to Your Content

The AI service must be enabled in your cloud provider settings before you can connect it to the platform. Only account owners can do this.

Once that is done:

1. Add a cloud account to your track.
2. Enable the relevant AI service within that account.

***

#### Limit Models and Token Usage

In a track, you can restrict which models learners can call and how many tokens each model can use.

1. Open the cloud account in the track's sandbox settings.
2. Enable Amazon Bedrock (AWS) or Vertex AI (Google Cloud) under **Services**.
3. Under **Bedrock Models** or **Vertex AI Models**, select the models learners can use. Leave the list empty to allow all supported models.
4. Optionally, set **Input tokens** and **Output tokens** for each selected model.

Learners can only call the selected models. Calls to any other model are denied.

{% hint style="info" %}
**Selecting models does not grant access to them.** The account still needs a policy that allows the AI service. For AWS, add a managed policy such as `arn:aws:iam::aws:policy/AmazonBedrockFullAccess`. Without it, every model call is denied, including calls to the selected models.
{% endhint %}

If a play goes over a model's token limit, the play is stopped.

You can also set the limits in `config.yml`:

```yaml
aws_accounts:
- name: aws
  services:
  - bedrock
  managed_policies:
  - arn:aws:iam::aws:policy/AmazonBedrockFullAccess
  service_limits:
    bedrock:
      models:
      - model: amazon.nova-micro-v1:0
        limits:
          input_token_count: <tokens>
          output_token_count: <tokens>
gcp_projects:
- name: gcp
  services:
  - aiplatform.googleapis.com
  service_limits:
    aiplatform.googleapis.com:
      models:
      - model: publishers/google/models/gemini-2.5-flash
        limits:
          input_token_count: <tokens>
```

***

#### Supported Models

Only text-based models are supported. Image and video generation models cannot be connected at this time.

The tables below list all currently supported models. If a model you need is not listed, see Missing models.

#### Google Vertex AI

{% hint style="info" %}
**Anthropic models are not available on Vertex AI.** Enabling Anthropic models in GCP requires completing a form in the Model Garden for each model. This process cannot be automated, so Anthropic models are not currently supported through Vertex AI. We recommend using AWS Bedrock for Anthropic models instead.
{% endhint %}

| Provider | Model                        |
| -------- | ---------------------------- |
| DeepSeek | deepseek-v4.1-flash-maas     |
| Google   | gemini-2.5-flash-lite        |
| Google   | gemini-2.5-flash             |
| Google   | gemini-2.5-pro               |
| Google   | gemini-3-flash-preview       |
| Google   | gemini-3.1-flash-lite        |
| Google   | gemini-3.1-pro-preview       |
| Google   | gemini-3.5-flash-lite        |
| Google   | gemini-3.5-flash             |
| Google   | gemini-3.6-flash             |
| Google   | gemini-3.7-flash             |
| Google   | gemini-3.8-flash             |
| Google   | gemini-4-argon               |
| Google   | gemma-4-26b-a4b-it-maas      |
| Meta     | llama-3.1-405b-instruct-maas |
| Mistral  | codestral-2                  |
| Mistral  | mistral-medium-3             |
| Mistral  | mistral-small-2503           |
| OpenAI   | gpt-oss-120b-maas            |

#### Amazon Bedrock

{% hint style="info" %}
**Anthropic models require all regions within a geography to be enabled.** Anthropic models on Bedrock use cross-region inference, which routes requests across all regions in a geography. You must enable model access in every region within the geography you want to use. For example, to use the US geography you need to enable all US regions: us-east-1, us-east-2, us-west-1, and us-west-2.

The first time an Anthropic model is used in a new sandbox, AWS finishes enabling it in the background and this can take a few minutes, as per [AWS documentation](https://docs.aws.amazon.com/bedrock/latest/userguide/model-access.html). Calls made during that time can fail with an access denied error mentioning AWS Marketplace. Retry after a few minutes.
{% endhint %}

| Provider  | Model                                       |
| --------- | ------------------------------------------- |
| Amazon    | nova-micro-v1:0                             |
| Amazon    | nova-lite-v1:0                              |
| Amazon    | nova-pro-v1:0                               |
| Amazon    | nova-2-lite-v1:0                            |
| Amazon    | titan-embed-text-v2:0                       |
| Anthropic | claude-haiku-4-5-20251001-v1:0              |
| Anthropic | claude-opus-4-5-20251101-v1:0               |
| Anthropic | claude-opus-4-6-v1                          |
| Anthropic | claude-opus-4-7                             |
| Anthropic | claude-opus-4-8                             |
| Anthropic | claude-opus-5                               |
| Anthropic | claude-opus-5-5 (Currently not supported)   |
| Anthropic | claude-sonnet-4-5-20250929-v1:0             |
| Anthropic | claude-sonnet-4-6                           |
| Anthropic | claude-sonnet-5 (Currently not supported)   |
| Anthropic | claude-sonnet-5-5 (Currently not supported) |
| DeepSeek  | v3-v1:0                                     |
| DeepSeek  | v3.2                                        |
| Google    | gemma-3-4b-it                               |
| Google    | gemma-3-12b-it                              |
| Google    | gemma-3-27b-it                              |
| Meta      | llama3-1-405b-instruct-v1:0                 |
| Meta      | llama3-3-70b-instruct-v1:0                  |
| Mistral   | devstral-2-123b                             |
| Mistral   | ministral-3-8b-instruct                     |
| Moonshot  | kimi-k3.1                                   |
| OpenAI    | gpt-oss-120b-1:0                            |

***

#### Requesting Additional Models

{% hint style="info" %}
**Missing models**

If the model you need isn't listed above, please reach out to the Instruqt Support team.

Every request kicks off a review on our side. We research whether the model is feasible to support and assess its security profile – including the guardrails and safety measures we'd need to put in place before enabling it. This research can take up to a month, after which we'll come back to you with our conclusion on feasibility and whether we'll be adding the model.
{% endhint %}

***

### FAQ

<details>

<summary>Which cloud providers are supported?</summary>

Google Vertex AI and Amazon Bedrock are currently supported. Additional providers may be added over time.

</details>

<details>

<summary>Can I use any model available in Vertex AI or Bedrock?</summary>

No. Only the models listed on this page are supported.

All other models are unavailable by default. If you need a model that is not listed, contact Instruqt Support. The review can take up to a month.

</details>

<details>

<summary>Who can connect a cloud account?</summary>

The AI service must be enabled in your cloud provider settings by an account owner or administrator before anyone can connect it on the platform.

</details>

<details>

<summary>Are there limits on how much I can use a model?</summary>

Yes. You can set input and output token limits per model on each cloud account in a track. See [Limit Models and Token Usage](#limit-models-and-token-usage).

</details>

<details>

<summary>Can I set custom limits for my organisation?</summary>

Limits are set per track, on each cloud account. Organisation-wide limits are not available.

</details>

<details>

<summary>Can I use image or video generation models?</summary>

No. Only text-based models are supported at this time. Image and video generation models cannot be connected.

</details>

<details>

<summary>What if the model I need is not on the supported list?</summary>

Contact Instruqt Support. We research whether the model is feasible to support and assess its security profile. This can take up to a month, after which we'll let you know whether we'll be adding the model.

</details>
