GitHub Models service retired

Summary

GitHub has completely discontinued its GitHub Models service. This service was a unified large language model (LLM) API and playground that allowed GitHub Actions workflows to call multiple model providers using an existing GitHub API key. Simon Willison noticed the shutdown when his own Actions pipeline broke and switched to a direct OpenAI key with limited spending.

Why it matters

GitHub Models was one of the few ways to run LLM requests in CI/CD pipelines without dealing with separate provider credentials. This made it a natural fit for GitHub's 'Continuous AI' vision. The removal forces developers to manage their own API keys, billing, and rate limits, eroding the frictionless integration that made the service appealing. The unannounced shutdown also suggests that subsidized multi-provider access is no longer economically sustainable for GitHub.

Highlights

• GitHub did not provide an official reason. Willison suggests that token costs for coding agents made free or subsidized access unsustainable.
• The service allowed Actions workflows to call multiple LLM providers with a single GitHub API key, requiring no additional credentials.
• Willison’s temporary solution: a direct OpenAI key with a monthly spending limit and the GPT-5.6 Luna model for README folder summaries.

Critical take

This is a thin link share. Willison offers no confirmation of the cost hypothesis, a timeline for the deprecation, or data on the number of affected projects. Therefore, the 'expensive token' explanation remains an unverified guess dressed up as analysis.