DeepSeek model on Azure: "This model is not supported by Responses API" error

[kyon1399]

Question

Hi team,

I'm using opencode cli as a daily development tool. I have deployed the DeepSeek-R1-0528 model on Azure and I can see the model listed in Opencode after the recent updates (when switching the provider to Azure). However, when I try to use DeepSeek-R1-0528 via Opencode using my Azure deployment, I get this error message:

This model is not supported by Responses API.

Here’s what I’ve tried so far:

1. I switched the provider from azure to azure-cognitive-services in Opencode, but the issue persists.
2. I attempted to add a custom provider configuration in opencode.json as shown below, but DeepSeek still does not work:

"azuredeepseek": { 
  "npm": "@ai-sdk/openai-compatible", 
  "name": "AzureDeepSeek", 
  "options": { 
    "baseURL": "https://XXXX.services.ai.azure.com/openai/v1/", 
    "apiKey": "XXXXX" 
  }, 
  "models": { 
    "DeepSeek-R1-0528": { 
      "name": "DeepSeek-R1-0528", 
      "limit": { 
        "context": 163840, 
        "output": 163840
      } 
    } 
  } 
}

Is anyone else experiencing similar issues?
Is there a workaround or specific configuration needed to make DeepSeek model work on Opencode via Azure?

Any advice or pointers would be greatly appreciated!

Thanks a lot!

[github-actions[bot]]

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

• #4943: github copilot: gpt-5-mini not supported via responses api
• #2967: GitHub Copilot (AI_APICallError: The requested model is not supported)

These issues appear to be related to the same "This model is not supported by Responses API" error, though with different models and providers. The root cause seems to be compatibility issues with the Responses API when using certain models.

Feel free to ignore if your specific case with DeepSeek-R1-0528 on Azure is distinct from these GitHub Copilot issues.

[rekram1-node]

@kyon1399 try this:

"azuredeepseek": { 
  "npm": "@ai-sdk/openai-compatible", 
  "name": "AzureDeepSeek", 
  "options": { 
    "baseURL": "https://XXXX.services.ai.azure.com/openai/v1/", 
    "apiKey": "XXXXX" 
  }, 
  "models": { 
    "DeepSeek-R1-0528": { 
      "name": "DeepSeek-R1-0528", 
      "limit": { 
        "context": 163840, 
        "output": 163840
      } ,
     "options": {"useCompletionUrls": true} // <<< this
    } 
  } 
}

[kyon1399]

@rekram1-node

Thank you for your reply. After trying your approach, I encountered another issue with my Azure DeepSeek configuration, even without adding the options.useCompletionUrls setting. The screenshots and logs are shown below. At first, I thought it was a rate‑limit problem, but the same DeepSeek R1 model works in my other AI clients, so I suspect it's not a limit issue. It might be that incorrect access settings are causing excessive requests.

ERROR 2025-12-05T02:42:13 +0ms service=session.processor error=Requests to the Creates a model response for the given chat conversation. Operation under Azure AI Model Inference API have exceeded token rate limit of your current AIServices S0 pricing tier. Please contact Azure support service if you would like to further increase the default rate limit. stack="AI_APICallError: Requests to the Creates a model response for the given chat conversation. Operation under Azure AI Model Inference API have exceeded token rate limit of your current AIServices S0 pricing tier. Please contact Azure support service if you would like to further increase the default rate limit.\n    at <anonymous> (../../node_modules/.bun/@[email protected]+d6123d32214422cb/node_modules/@ai-sdk/provider-utils/dist/index.mjs:889:18)\n    at processTicksAndRejections (native:7:39)" process

[rekram1-node]

now thats surprising whats the ratelimit

[kyon1399]

I have a rate limit of 20,000 tokens per minute and 20 requests per minute with the DeepSeek R1 model. At first I thought that because opencode showed I had exceeded the rate while another AI client could still access the R1 model, the problem wasn’t a rate limit. But I tried calling the MAI-DS-R1 model from opencode (which has a rate limit of 5 million tokens per minute and 5,000 requests per minute), and opencode returned a response successfully. I believe my issue is resolved — thank you for your help!

[rekram1-node]

no problem