Azure AI Foundry/Kimi K2.5: "This model is not supported by the Responses API"

Description

We deployed Kimi K2.5 to Azure AI Foundry, set the environment variables according to the docs and connected with the API key thru the TUI. Every request to the Kimi K2.5 model fails though with the error message.

Plugins

none

OpenCode version

1.1.53

Steps to reproduce

Deploy Kimi K2.5 to Azure AI foundry
Set up AZURE_COGNITIVE_SERVICES_RESOURCE_NAME as described in the docs
Set up opencode.json:

{
    "$schema": "https://opencode.ai/config.json",
    "autoupdate": true,
    "provider": {
        "azure-cognitive-services": {
            "options": {
                "apiKey": "<redacted>",
                "resourceName": "<redacted>"
            },
            "models": {
                "Kimi-K2.5": {
                    "name": "Kimi-K2.5-az"
                }
            }
        }
    }
}

auth opencode via opencode auth login oder /connect
Select the model and send any message

Screenshot and/or share link

No response

Operating System

Windows 11

Terminal

Windows Terminal

This issue might be a duplicate of existing issues. Please check:

Bad request when using Kimi K2.5 on Azure Foundry #12879: Bad request when using Kimi K2.5 on Azure Foundry (same platform/model combination with a related error)
kimi-k2.5 model not available (Model not found/inaccessible/not deployed) #12917: kimi-k2.5 model not available (Model not found/inaccessible/not deployed)

I think that this may be that the Foundry Base Models use a different endpoint from the standard models. I got my Kimi-K2.5 to respond to the text (with quota exceeded I must admit) with setting the endpoint manually.
Here is the respective part of my config:

    "provider": {
        "azure-foundry-base-models": {
            "name": "Azure Foundry Base Models",
            "npm": "@ai-sdk/openai-compatible",
            "options": {
                "baseURL": "https://swedencentral.api.cognitive.microsoft.com/openai/v1"
            },
            "models": {
                "Kimi-K2.5": {
                    "name": "Kimi-K2.5"
                }
            }
        }
    },

You can find the endpoint on the page of the model deployment in the code example.

I cannot say however, why I have the quota exceeded issue, since I when using curl I get the answer just fine:

$  curl -s "https://swedencentral.api.cognitive.microsoft.com/openai/v1/chat/completions" \                                                                                                                                                                           
         -H "Authorization: Bearer $AZURE_API_KEY" \
         -H "Content-Type: application/json" \
         -d '{"model": "Kimi-K2.5", "messages": [{"role": "user", "content": "Tell me hi!"}]}'
{"choices":[{"content_filter_results":{"hate":{"filtered":false,"severity":"safe"},"protected_material_code":{"filtered":false,"detected":false},"protected_material_text":{"filtered":false,"detected":false},"self_harm":{"filtered":false,"severity":"safe"},"sexual":{"filtered":false,"severity":"safe"},"violence":{"filtered":false,"severity":"safe"}},"finish_reason":"stop","index":0,"logprobs":null,"message":{"content":"Hi! 👋 How can I help you today?","reasoning_content":"The user is asking me to say \"hi\" to them. This is a very simple request. I should respond in a friendly and welcoming manner.\n\nSince this is just a greeting and there's no complex task or potential harm involved, I can simply respond with a greeting.\n\nI'll say \"Hi!\" or something similar like \"Hello!\" or \"Hi there!\" to keep it friendly and appropriate. ","role":"assistant","tool_calls":null},"stop_reason":null}],"created":1771541816,"id":"0c26a35da4c140b3b5107072eb893ee0","model":"Kimi-K2.5","object":"chat.completion","prompt_filter_results":[{"prompt_index":0,"content_filter_results":{"hate":{"filtered":false,"severity":"safe"},"jailbreak":{"filtered":false,"detected":false},"self_harm":{"filtered":false,"severity":"safe"},"sexual":{"filtered":false,"severity":"safe"},"violence":{"filtered":false,"severity":"safe"}}}],"usage":{"audio_prompt_tokens":0,"completion_tokens":91,"prompt_tokens":12,"prompt_tokens_details":null,"reasoning_tokens":0,"total_tokens":103}}

Can anyone comment on that?
Is this method working for you?

Can confirm that this fixes the responses API error, the endpoint URL from the playground code sample seems to work. However, I get the same quota error that you reported while the chat playground doesn't bring up any quota errors.

@gdl-reif

@gdl-reif and do you have any ideas on what that can be caused by? You have sufficient quota? It does seem as a Azure issue, right?

@kunzaatko

@kunzaatko frankly speaking - no idea. The code samples and playground in Azure work as expected within our assigned quotas, but I get the quota error using the same model deployment when using it with OpenCode. From my tests, it looks more like a bug in OpenCode as the model can be used from the sample code.

To me as well since the curl works just fine. Good to know that it is not an individual issue and misconfiguration. That gives some direction at least

@gdl-reif

@gdl-reif @kunzaatko to fix that error you need to reduce the output for that model like so.

        "kimi-k2.5": {
          "id": "Kimi-K2.5",
          "limit": {
            "output": 10000, // Reduce limit.output for models where you get rate-limit errors
            "context": 262144
          }
        }

You can play around with the one they accept

@OpeOginni

@OpeOginni thanks - the rate limit is not (yet) the issue in OpenCode, currently the model is completely unsuable due to the error mentioned in the title "This model is not supported by the Responses API"

@gdl-reif

@gdl-reif got a fix for that, you would need to move None openAI models off the azure provider, and login to the azure cognition services using opencode auth login and make use of this config

    "azure-cognitive-services": {
      "npm": "@ai-sdk/openai-compatible", 
      "options": {
        "baseURL": "https://{resource_name}.openai.azure.com/openai/v1",
        "useCompletionUrls": true
      },
      "models": {
        "kimi-k2.5": {
          "id": "Kimi-K2.5",
          "limit": { "output": 10000, "context": 262144 }
        }
      }
    }

@OpeOginni

@OpeOginni can confirm that this configuration works with my Azure Foundry deployment of Kimi-K2.5 in the West Germany region - I did some quick (!) tests and did neither run into the Responses API error nor into a resource limit/rate limit.

@gdl-reif

Nice to hear @gdl-reif hopefully the team approves the PR that adds the better explanation to the docs

@gdl-reif @kunzaatko to fix that error you need to reduce the output for that model like so.
    "kimi-k2.5": {
      "id": "Kimi-K2.5",
      "limit": {
        "output": 10000, // Reduce limit.output for models where you get rate-limit errors
        "context": 262144
      }
    }

I can confirm that this works on my machine... Thank you!

What I still do not understand though is the baseURL that I need to pass. You are using the Azure models with the "https://{resource_name}.openai.azure.com/openai/v1" endpoint? When I open the Foundry page it shows the target URI as https://swedencentral.api.cognitive.microsoft.com/models/chat/completions?api-version=2024-05-01-preview, which does not translate to the template with the resource_name that is the default. What is the problem there? Is that for the older version of Azure provider? Here is a screenshot from the Foundry portal:

You can find the correct URL in the playground sample code:

@gdl-reif

@gdl-reif And you are using the old or the new Foundry portal? I have only this:

The endpoint that I provided works, but I have to configure the endpoint directly using the "@ai-sdk/openai-compatible" and "baseURL" so I was wondering, whether I could plug into the default Azure opencode provider configuration only when using the new Foundry portal...

@gdl-reif

@gdl-reif And you are using the old or the new Foundry portal? I have only this:

The endpoint that I provided works, but I have to configure the endpoint directly using the "@ai-sdk/openai-compatible" and "baseURL" so I was wondering, whether I could plug into the default Azure opencode provider configuration only when using the new Foundry portal...

That azure provider only wild with open ai models, hence why it's best to make a separate provider you can put in all your non OpenAI models that are OpenAI compatible like Kimi

@gdl-reif And you are using the old or the new Foundry portal? I have only this:

I'm on the new "experience", but I remember the playground being there in the old version as well.

For anyone who ends up here and doesn't understand why the endpoint "https://{resource_name}.openai.azure.com/openai/v1" is not available to them, the custom_subdomain_name property of the foundry resource must be set up either in the portal or using terraform https://registry.terraform.io/providers/hashicorp/azurerm/latest/docs/resources/cognitive_account.

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

Navigation Menu

Repository navigation

Azure AI Foundry/Kimi K2.5: "This model is not supported by the Responses API" #12949

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Azure AI Foundry/Kimi K2.5: "This model is not supported by the Responses API" #12949

Description

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

github-actions commented on Feb 10, 2026

kunzaatko commented on Feb 19, 2026

gdl-reif commented on Feb 20, 2026

kunzaatko commented on Feb 20, 2026

gdl-reif commented on Feb 20, 2026

kunzaatko commented on Feb 20, 2026

OpeOginni commented on Feb 21, 2026

OpeOginni commented on Feb 21, 2026

gdl-reif commented on Feb 23, 2026

OpeOginni commented on Feb 23, 2026

gdl-reif commented on Feb 23, 2026

OpeOginni commented on Feb 23, 2026

kunzaatko commented on Feb 24, 2026

kunzaatko commented on Feb 24, 2026

gdl-reif commented on Feb 24, 2026

kunzaatko commented on Feb 24, 2026

OpeOginni commented on Feb 24, 2026

gdl-reif commented on Feb 24, 2026

kunzaatko commented on Feb 24, 2026

github-actions commented on Apr 26, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions