ai-sdk/azure calling the wrong url path for r1 deployment #4797

Activity

1. sahanatvessel

   added

   bugSomething isn't working as documentedSomething isn't working as documented

   on Feb 9, 2025
2. kevsjh commented on Feb 13, 2025

   

   kevsjh

   on Feb 13, 2025

   More actions

   Yes, im facing the same issue as well. Applying baseURL does not fix it
3. Genzo0 commented on Feb 17, 2025

   

   Genzo0

   on Feb 17, 2025

   More actions

   Does anyone find the solution for this problem ?
4. sahanatvessel commented on Mar 3, 2025

   

   sahanatvessel

   on Mar 3, 2025

   Author

   More actions

   @lgrammel any insights on this, please?

   👍React with 👍1Reacted by Tobias Büschel
5. tobiasbueschel commented on Mar 19, 2025

   

   tobiasbueschel

   on Mar 19, 2025

   Contributor

   More actions

   I've been having a similar issue when using a custom gateway to access Azure hosted LLMs. This is the URL that @ai-sdk/azure is creating after I've modified the baseURL, which is incorrect:

   https://{GATEWAY}.com/llama2-70b-chat/{CLIENT_ID}/v1/llama-2-70b-chat/chat/completions?api-version=2024-10-01-preview
   

   This is the actual URL I need for my gateway:

   https://{GATEWAY}.com/llama2-70b-chat/{CLIENT_ID}/v1/chat/completions
   

   I've tried using a custom fetch implementation in an attempt to correct the URL, but passing a newUrl to the return fetch call does not change where the request is sent to and is also not a great way to fix this issue:

   const llama2_70b_chat = createAzure({
     apiKey: AI_GATEWAY_KEY,
     baseURL: `https://${GATEWAY}.com/llama2-70b-chat/${CLIENT_ID}/v1`,
     fetch: async (url, options) => {
       const newUrl = url.replace("v1/llama-2-70b-chat", "v1");
       return await fetch(newUrl, options);
     },
   });

   Has anyone found a way to solve the above? Or should one use a custom provider for situation like this?

   Thanks

   👍React with 👍1Reacted by Ben Kissi
6. Genzo0 commented on Mar 19, 2025

   

   Genzo0

   on Mar 19, 2025

   More actions

   its keep the openai completions in the url, for know i think its better to look at the other provider thats support your model
7. lgrammel

   added

   provider/azureIssues related to the @ai-sdk/azure providerIssues related to the @ai-sdk/azure provider

   on Jun 26, 2025
8. aayush-kapoor commented on Jan 22, 2026

   

   aayush-kapoor

   on Jan 22, 2026

   Collaborator

   More actions

   Marking this as closed for now - if you face any regression or have a similar problem after migrating to the latest AI SDK version, please feel free to open a new issue (with possible reproduction)
9. aayush-kapoor

   closed this as completedon Jan 22, 2026
10. lgrammel commented on Jul 22, 2026

    

    lgrammel

    on Jul 22, 2026

    Collaborator

    More actions

    Identify issue type

    Identified as Bug
    Confidence: High

    Rationale

    This describes concrete incorrect request URL construction in the existing Azure provider, causing requests to target endpoints different from those configured or required.

    Evidence

    • The reporter states that @ai-sdk/azure constructs the wrong URL for an Azure-hosted R1 deployment by adding an OpenAI-specific path.
    • Another user provides concrete generated and expected URLs, showing that the SDK inserts an unwanted model/deployment segment and api-version query even when baseURL is customized.
    • Multiple commenters confirm the same behavior, and changing baseURL reportedly does not resolve it.
11. lgrammel

    added

    task-identify-issue-type-done

    on Jul 22, 2026
12. lgrammel

    added the

    Bug

    issue type on Jul 22, 2026