Azure Foundry for opencode

[kevinzhg]

Question

I am using the below in opencode.json,
"azure": {
"npm": "@ai-sdk/anthropic",
"options": {
"apiKey": [mykeys],
"baseURL": "https://my-resource.services.ai.azure.com/anthropic/v1/messages"
},
"models": {
"claude-opus-4-6": {
"name": "claude-opus-4-6"
}
}
}
And i got the error below, do you happen to know whats the problem and how to fix it?

hello
Not Found: {"error":{"code":"api_not_supported","message":"Requested API is currently not supported","details":"Requested API is currently not supported"}}
▣ Build · claude-opus-4-6

[rekram1-node]

Why do u need a custom provider for this??

[WillySagzz]

Why do u need a custom provider for this??

Could be because it is a deployment inside an enterprise or company tenant.
I've seen that case with some companies that would rather deploy their own endpoints so that IT can monitor the usage.

[pandma]

Hi 👋 — opened #28073 which partially addresses this for OpenAI-compatible models on Foundry (gpt-5.x etc.). It adds:

• A /login flow for foundry with prompts for resource hostname, project name, and deployment names
• A chat.params hook that strips the params Foundry's OpenAI endpoint rejects (reasoningSummary, reasoningEffort, textVerbosity, max_tokens capping) so gpt-5.x deployments work out of the box

Your specific case (Claude on Foundry via /anthropic/v1/messages) isn't covered yet — that needs @ai-sdk/anthropic and a different URL pattern, which would be a clean follow-up on top of this PR. Happy to take that on if the first one lands.