AIBRIEF
GitHub's free LLM playground and unified API, usable from any GitHub Actions workflow with the existing token, was retired on 9 August 2026 with no public postmortem.
GitHub retired GitHub Models on 9 August 2026. The product, a free model playground and unified API, let code in GitHub Actions call any large language model using the GitHub token already in the environment.
Most LLM access needs a separate provider key, a billing relationship, and per-call pricing. GitHub Models sat inside the same authentication every Actions workflow already had, which made it a natural fit for GitHub Next's "Continuous AI" pitch: model calls as a routine pipeline step, alongside test runs and deploy hooks. The retirement removes that pattern at the source.
Simon Willison, who relied on it to generate folder summaries in a README workflow, first surfaced the shutdown via a failed Actions run on his simonw/research repository. The Actions log returned a brownout error message that has since gone stale, and GitHub has not, in the source, published an official reason for the shutdown.
Willison attributes the shutdown to coding-agent economics: subsidized or free token access became prohibitively expensive once agents started burning tokens at scale. That is his bet, not GitHub's.
His own migration, an OpenAI API key with a hard monthly spending limit pinned to GPT-5.6 Luna, is how he now runs AI calls inside GitHub Actions.