Getting "PermissionDeniedError: 403 No access to model: openai/o4-mini

[sathwik-3721]

Body

Hey all!!

I have been trying to access GitHub models through code, I have created my GitHub PAT and accessing the models. But while accessing them through code in my machine, Getting this error

Getting "PermissionDeniedError: 403 No access to model: openai/o4-mini while accessing GitHub Models via code through my machine

Could any one please help me out in solving this issue.

Thanks in advance!!

[thevibingteen]

That PermissionDeniedError: 403 No access to model: openai/o4-mini means your token/permissions or model access settings don’t allow sampling that model. Common causes and how to fix them:

Wrong API / endpoint / token type — if you’re trying to call GitHub Models you must call the GitHub Models endpoint (https://models.github.ai/...) and authenticate with a GitHub PAT that has the models scope. If you instead call the OpenAI endpoint with a GitHub PAT (or vice versa) you’ll get 403s. See the GitHub Models quickstart for the correct curl example.
GitHub Docs
+1

PAT scope / fine-grained token permissions — verify your PAT includes the models scope (or models:read for fine-grained tokens). Create a new PAT with the right scope and try again.
GitHub Docs
+1

Model not enabled for your account / org / plan — o4-mini may require you to enable that model in your Copilot/GitHub settings or be on a plan that exposes the model. For personal use you may be prompted to “Allow” a model the first time you pick it; for org/enterprise the administrator must enable the model in Copilot policies. Ask your org admin to enable o4-mini if you’re in an organization.
GitHub Docs
+1

Plan / usage restrictions — some models are gated to particular Copilot plans or usage tiers. Confirm your account/plan includes o4-mini access. If you hit a rate/usage limit the API might refuse sampling.
The GitHub Blog

If everything looks correct — debug steps: (a) regenerate a new PAT with models scope and test a minimal curl (below); (b) call a simple listing or a known working model (e.g., openai/gpt-4.1) to confirm auth; (c) capture the full 403 response body (it often contains a reason like “model not enabled for org”); (d) if org-managed, ask your org admin or open a GitHub support ticket with the error + request.
GitHub Docs
+1

Minimal curl test (copy/paste and replace YOUR_GITHUB_PAT)

bash
Copy
Edit
curl -L -X POST
-H "Accept: application/vnd.github+json"
-H "Authorization: Bearer YOUR_GITHUB_PAT"
-H "X-GitHub-Api-Version: 2022-11-28"
-H "Content-Type: application/json"
https://models.github.ai/inference/chat/completions
-d '{"model":"openai/o4-mini","messages":[{"role":"user","content":"Say hi"}]}'
If that returns 403, the response body usually states whether it’s a permission/plan/organization policy issue — paste that into the thread if you want help reading it.

TL;DR — check (A) you’re using the GitHub Models endpoint, (B) your PAT has models scope, and (C) the model is enabled for your account/org/plan. If all three are correct and you still get 403, capture the response body and contact GitHub support or your org admin.

[thevibingteen]

sure
if it helps..then mark it as an answer so that it may help others too!

[sathwik-3721]

Hey @thevibingteen, Got this in Postman for the curl

Also could you please help me how to find this "enable that model in your Copilot/GitHub settings"

[thevibingteen]

Hi @sathwik-3721 — this 403 is very common. It means either your token, endpoint, or account/org settings don’t allow sampling openai/o4-mini. Try these steps (I included a minimal curl you can run in Postman or terminal):

1. Use the GitHub Models endpoint and correct headers
   Make sure your request goes to https://models.github.ai/... and includes:

Authorization: Bearer <YOUR_GITHUB_PAT>

Accept: application/vnd.github+json

X-GitHub-Api-Version: 2022-11-28

Content-Type: application/json

2. Ensure your PAT has the models scope

If using a classic PAT, enable the models scope.

If using a fine-grained token, ensure it includes models:read (or the equivalent models permission).

3. Confirm the model is enabled for your account/org

Personal account: Go to GitHub → click your profile picture → Your Copilot (or Copilot settings). If o4-mini is blocked you should see an “Allow / Enable” prompt — accept it.

Organization: Ask an org owner/Copilot admin to enable the model: Org → Settings → Copilot → Policies / Models and allow o4-mini for the org or specific teams/users.

4. Quick minimal curl (copy/paste, replace token)

curl -L -X POST
-H "Accept: application/vnd.github+json"
-H "Authorization: Bearer YOUR_GITHUB_PAT"
-H "X-GitHub-Api-Version: 2022-11-28"
-H "Content-Type: application/json"
https://models.github.ai/inference/chat/completions
-d '{"model":"openai/o4-mini","messages":[{"role":"user","content":"Say hi"}]}'

Run that from your machine / Postman. If it still returns 403, the JSON response body often includes the reason (e.g., "model not enabled for org", or "token missing models scope").

5. If everything looks correct

Regenerate a new PAT with the models scope and retry.

Confirm you’re not accidentally calling the OpenAI endpoint (or using an OpenAI key) — endpoint + token type must match.

[sathwik-3721]

Hi @thevibingteen I have been using this module "Azure AI Inference SDK". not OpenAI, And the endpoint I'm using is https://models.github.ai/inference/chat/completions
Also when I run the curl, got this

{
    "error": {
        "code": "unavailable_model",
        "message": "Unavailable model: o4-mini",
        "details": "Unavailable model: o4-mini"
    }
}

Here is what my copilot settings looks like.

[thevibingteen]

Hi @sathwik-3721 — thanks for the follow-up and the Postman output. That JSON:

{"error":{"code":"unavailable_model","message":"Unavailable model: o4-mini","details":"Unavailable model: o4-mini"}}

means the model itself isn’t available to your account/org (so your endpoint and token flow look correct). A couple of quick notes and exact steps an admin can take to fix this:

Quick note

The Azure AI Inference SDK talks to Azure endpoints; using it can accidentally target the wrong service. Calling https://models.github.ai/... via direct HTTP (curl/Postman) as you did is the right way to test GitHub Models.

What an admin should do (quick)

Go to Organization → Settings → Copilot → Policies / Models and allow openai/o4-mini for the org or the affected team/user.

Confirm the org’s Copilot plan includes access to o4-mini (some models are plan-gated).

Check any model allow/deny lists and ensure there’s no deny rule blocking o4-mini.

Commands to run (paste these)

List models visible to the PAT

curl -H "Accept: application/vnd.github+json"
-H "Authorization: Bearer YOUR_GITHUB_PAT"
-H "X-GitHub-Api-Version: 2022-11-28"
https://models.github.ai/models

Expected result after enabling: the output should include openai/o4-mini.

Test a simple inference

curl -X POST
-H "Accept: application/vnd.github+json"
-H "Authorization: Bearer YOUR_GITHUB_PAT"
-H "X-GitHub-Api-Version: 2022-11-28"
-H "Content-Type: application/json"
https://models.github.ai/inference/chat/completions
-d '{"model":"openai/o4-mini","messages":[{"role":"user","content":"Say hi"}]}'

Expected: a normal completion JSON (not 403 / unavailable_model).

Extra checks for the requester

Make sure the PAT used for testing has models scope (classic) or models:read (fine-grained).

If the model still doesn’t appear after an admin enables it, paste the full JSON returned by the GET /models call here — that response usually contains the exact policy/plan reason and I’ll help interpret it.

[mt-abdullah]

Hello sathwik-3721,

The PermissionDeniedError: 403 No access to model: openai/o4-mini error typically occurs when your GitHub Personal Access Token (PAT) lacks the necessary permissions to access the specified model or repository. Here are some steps to troubleshoot and resolve the issue:

1. Verify PAT Permissions:

   • Ensure your GitHub PAT has the required scopes. For accessing models, you may need read:packages or other relevant permissions. You can check and update your PAT scopes in your GitHub account settings.

2. Check Repository Visibility:

   • Confirm that the model repository (openai/o4-mini) is public or that your GitHub account has been granted access if it’s private. If it’s a private repository, you may need explicit access from the owner.

3. Review Code Implementation:

   • Double-check how you’re using the PAT in your code. Ensure the token is correctly passed in the request headers. For example:

     headers = {
         "Authorization": f"Bearer YOUR_GITHUB_PAT",
         "Accept": "application/vnd.github.v3+json"
     }

4. Test Access via API:

   • Try accessing the repository or model directly via the GitHub API to verify permissions. For example:

     curl -H "Authorization: Bearer YOUR_GITHUB_PAT" https://api.github.com/repos/openai/o4-mini
     

   • If this fails, the issue is likely with the PAT or permissions.

5. Contact Repository Owner:

   • If the model is private and you believe you should have access, reach out to the repository owner or maintainers to confirm your access rights.

6. Debugging Tools:

   • Use tools like gh auth status (GitHub CLI) or browser developer tools to inspect network requests for additional error details.

If you’ve checked all the above and the issue persists, please share more details about your code and how you’re attempting to access the model. This will help in diagnosing the problem further.

Let me know if you need additional assistance!

[DanielRizvi]

A PermissionDeniedError: 403 No access to model: openai/o4-mini means your token is valid but the account or token you’re using does not have permission to use that specific model on GitHub Models.

A few things to check:

Model availability

Not all models on GitHub Models are publicly accessible — some require you to be in a beta program or have been granted access to the publisher’s model.

openai/o4-mini might be in limited preview. You’ll need to confirm it’s listed as “available” in your GitHub Models dashboard.

Personal Access Token (PAT) scope

For GitHub Models API, the token must have the read:models scope.

Create or update your token in Settings → Developer settings → Personal access tokens and make sure you select the correct scope.

Correct endpoint and authentication
Example Python snippet:

import requests

headers = {
"Authorization": "Bearer <YOUR_PAT>",
"Content-Type": "application/json"
}

url = "https://api.github.com/models/openai/o4-mini/infer"
data = {"input": "Hello!"}

response = requests.post(url, headers=headers, json=data)
print(response.text)

Make sure your endpoint matches the model’s documentation.

Account type

If you’re using a work account, your org admin might need to enable access to external models.

Test from GitHub Codespaces

If it works in Codespaces but not locally, the issue might be firewall/proxy restrictions or local token environment variable setup.

In short:

Check if openai/o4-mini is available to your account.

Ensure your PAT has read:models scope.

Verify you’re calling the correct endpoint with proper headers.

If you confirm all of the above and still get a 403, you’ll need to contact GitHub Support to check your account’s model access.

[abhivarma007]

I am trying to use actions/ai-inference@v1 action in my workflow and getting the error

Error: 403 No access to model: /claude-3.5-sonnet do anyone know how to fix it?

on:
  workflow_dispatch

permissions:
  contents: read
  models: read
  
jobs:
  inference:
   runs-on: ubuntu-latest
   steps:
     - uses: actions/checkout@v4

     - uses: actions/ai-inference@v1
       with:
         prompt: 'Hi'
         enable-github-mcp: false
         token: ${{ github.token }}
         github-mcp-token: ${{ secrets.MCP_**_TOKEN }}

[DanielRizvi]

GitHub Actions Workflow for AI Inference

This workflow runs an AI inference job using the actions/ai-inference@v1 action.

name: AI Inference Workflow

on:
  workflow_dispatch:

jobs:
  inference:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      models: read

    steps:
      - uses: actions/checkout@v5

      - uses: actions/ai-inference@v1
        with:
          prompt: "Hi"
          model: "anthropic/claude-3-5-sonnet-latest"
          enable-github-mcp: false
          token: ${{ github.token }}
          github-mcp-token: ${{ secrets.MCP_**_TOKEN }}

[Ryan-Palmer]

My PAT worked for personal use but when I created a PAT owned by my org (so I could use their billing) it stopped working. After a lot of searching I realised the endpoint is different for orgs, i.e.
https://models.github.ai/orgs/<org name>/inference
instead of the usual
https://models.github.ai/inference

https://docs.github.com/en/rest/models/inference?apiVersion=2022-11-28#run-an-inference-request-attributed-to-an-organization

It would be great to have this called out more explicitly in the guides.