Voice Agent managed Anthropic models fail with INVALID_SETTINGS: model not available

[myswitchboard-io]

Hey Deepgram team - I’m trying to get Anthropic Haiku working through Deepgram Voice Agent, and I’m looking for help confirming the correct model ID or provisioning path.

We specifically want to use Haiku as the agent.think model for our Voice Agent setup. OpenAI is only our temporary fallback because it is the only thing currently getting the session through settings-applied.

What is happening

We are using the Deepgram Voice Agent SDK (@deepgram/agents).

When we configure agent.think with OpenAI, the session connects and the call works.

When we configure agent.think with Anthropic Haiku, the session fails during setup with:

{
  "code": "INVALID_SETTINGS",
  "description": "Invalid agent.think settings - model not available",
  "type": "Error"
}

After that, the SDK retries with exponential backoff, but it appears to keep retrying the same rejected settings. The session never reaches settings-applied, and no audio plays.

Haiku model IDs we tried

First we tried the Haiku model ID listed in the Voice Agent LLM Models docs:

agent.think = {
  provider: {
    type: "anthropic",
    model: "claude-4-5-haiku-latest"
  },
  prompt: "<assembled caller system prompt>"
}

That failed with:

INVALID_SETTINGS: Invalid agent.think settings - model not available

Then we tried the alternate Haiku 4.5 ID mentioned in the Voice Agent prompting docs:

agent.think = {
  provider: {
    type: "anthropic",
    model: "claude-haiku-4-5"
  },
  prompt: "<assembled caller system prompt>"
}

That also failed with the same error:

INVALID_SETTINGS: Invalid agent.think settings - model not available

So the issue does not appear to be only the claude-4-5-haiku-latest alias.

The rest of our settings object

The rest of the settings object is unchanged between the failing Haiku tests and the working OpenAI test.

agent.listen.provider = {
  type: "deepgram",
  version: "v2",
  model: "flux-general-en",
  eot_threshold: 0.75
}

agent.speak.provider = {
  type: "deepgram",
  model: "aura-2-asteria-en"
}

agent.language = "en"

Authentication is handled with a short-lived Deepgram token minted through /v1/auth/grant with a 60-second TTL.

OpenAI works with the same setup

If we change only the agent.think.provider block to OpenAI, the call works normally:

agent.think = {
  provider: {
    type: "open_ai",
    model: "gpt-4o-mini"
  },
  prompt: "<same assembled caller system prompt>"
}

With OpenAI:

• The session reaches settings-applied
• Audio plays end-to-end
• No INVALID_SETTINGS error is thrown
• The call runs normally

So from what we can tell, this likely rules out our auth flow, proxy assembly, prompt formatting, audio I/O settings, SDK setup, and network/CORS configuration.

The only meaningful change between the working and failing versions is:

agent.think.provider.type
agent.think.provider.model

Why this is confusing

The Voice Agent LLM Models docs list Anthropic as a managed provider and include claude-4-5-haiku-latest.

The Voice Agent prompting docs appear to mention that claude-4-5-haiku-latest may return INVALID_SETTINGS, and suggest claude-haiku-4-5 instead.

In our project, both of those Haiku IDs fail.

What we need help with

Can someone from Deepgram confirm the current correct way to use Anthropic Haiku with Voice Agent?

Specifically:

1. Is managed Anthropic Haiku available to all Voice Agent projects, or does it require account/project-level enablement?

2. Is there a support-side provisioning step, dashboard setting, account tier, or add-on required before managed Anthropic models work?

3. What exact model ID should we use for Anthropic Haiku 4.5 in agent.think.provider.model?

4. Should the correct ID be claude-4-5-haiku-latest, claude-haiku-4-5, or something else?

5. Does the runtime validator use the same model registry as the public docs and the /v1/agent/settings/think/models endpoint?

6. If Haiku is not available to our project yet, is INVALID_SETTINGS: model not available the expected error?

Current status

Our temporary fallback is:

agent.think.provider = {
  type: "open_ai",
  model: "gpt-4o-mini"
}

That works, but it is not the desired end state.

The desired end state is to use managed Anthropic Haiku through Deepgram Voice Agent. We mainly need to understand whether this is:

• the wrong Haiku model ID,
• a docs/runtime model registry mismatch,
• or an account/project provisioning issue.

Thanks — happy to provide any additional settings payload details if useful.

[deepgram-community[bot]]

Thanks for asking your question. Please be sure to reply with as much detail as possible so the community can assist you efficiently.
_{Consider joining our Discord community for more opportunity to engage with your fellow Deepgram users. You can earn points which can be redeemed for cool stuff by being active in our communities!}

[deepgram-community[bot]]

Hey there! It looks like you haven't connected your GitHub account to your Deepgram account. You can do this at https://community.deepgram.com - being verified through this process will allow our team to help you in a much more streamlined fashion.

[deepgram-community[bot]]

It looks like we're missing some important information to help debug your issue. Would you mind providing us with the following details in a reply?

• The programming language you are working in (e.g. JavaScript, Python).
• A request ID that triggered your error or issue.