Pattern: GitHub Models Endpoint Migration (Azure → github.ai) | AI Memex

Discovery

While validating a deep research report for a Markdown-to-Linked-Data knowledge bank project, we found the report referenced https://models.inference.ai.azure.com as the GitHub Models API endpoint. This endpoint was deprecated in July 2025 and stopped responding entirely in October 2025.

The report also used bare model names like "gpt-4o-mini" — these now 404 because the API requires publisher/model format.

Root Cause

GitHub Models was initially hosted on Azure infrastructure and used Azure-style model naming. In mid-2025, GitHub migrated to its own domain (models.github.ai) with a new naming convention. This is a common pattern with platform services — initial launches use shared infrastructure before graduating to dedicated endpoints.

The change is documented in the GitHub Blog changelog but is easy to miss if you're working from older tutorials, cached documentation, or AI-generated research reports that trained on pre-migration data.

Solution

Endpoint: https://models.github.ai/inference

Model naming: Use publisher/model format:

openai/gpt-4o-mini (not gpt-4o-mini)
openai/gpt-4o (not gpt-4o)

Python SDK (OpenAI-compatible):

from openai import OpenAI
import os

client = OpenAI(
    base_url="https://models.github.ai/inference",
    api_key=os.environ["GITHUB_TOKEN"]
)

response = client.chat.completions.create(
    model="openai/gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello"}],
    temperature=0
)

Sources:

Prevention

Always check the current GitHub Models documentation before starting a project — endpoints may have changed
AI-generated research reports (including from deep research tools) may reference stale endpoints — validate API URLs independently
Pin the endpoint URL in a single config constant so migration is a one-line change