What Happened
GitHub will remove six models from Copilot Chat, edits, agent modes, and completions on December 14, 2026. It recommends successors for five; a replacement for Kimi K3 has not been announced. Claude Haiku 4.5 remains available to individual subscribers on annual plans. Business and Enterprise administrators should check whether suggested alternatives are permitted by their model policies [1].
Gemini 4 Hydrogen is rolling out as a generally available Copilot option across supported IDEs and GitHub surfaces. GitHub describes it as suited to fast, multistep coding and tool use; organizations can control access through model policies [3]. The Copilot app also now supports GitHub Enterprise Server accounts for remote repositories, issues, and pull requests, although cloud sessions are unavailable for those accounts [2].
A public-preview REST API setting lets organizations script repository AI Scan policy. Enabling it does not run a scan: a user with write access must start one, and each scan consumes AI credits. Enterprise policy can prevent enablement, and GitHub Enterprise Server is not supported [4]. The available release details establish these GitHub changes; they do not establish new releases for Cursor, Windsurf, Replit, Sourcegraph, JetBrains, VS Code, or Continue.dev.
Why It Matters to Businesses
A model retirement can change coding-assistant behavior without a change to application code. Teams that pin models in instructions, integrations, or evaluations need to test replacements before the deadline, especially for agent workflows that invoke tools. The new scan API makes security rollout easier to automate, but policy enablement and scan execution remain separate decisions [1][4].
Kimbodo Engineering Perspective
We would treat model selection as a governed dependency, not a developer preference alone. A faster model may improve interaction time while changing code quality, tool-use behavior, and spend; GitHub’s positioning of Gemini 4 Hydrogen is a reason to benchmark it, not evidence that it will improve a particular team’s results [3]. Likewise, GHES account support does not imply parity with cloud-connected Copilot sessions [2].
How We Would Implement It
- Inventory Copilot model selections and any integrations that specify the six retiring models. Test supported replacements against representative coding and agent tasks, then update policies and instructions before December 14 [1].
- Pilot Gemini 4 Hydrogen with a bounded group. Compare task completion, review corrections, latency, and billed usage against the current approved model before widening access [3].
- For GHES users, deploy Copilot app version 1.2.0 or later, connect a GitHub.com or GHE.com account first, and verify repository, issue, and pull-request workflows without relying on cloud sessions [2].
- Roll out AI Scan through the REST API in stages. Check enterprise policy and licensing first, set organization defaults and repository exceptions, then schedule explicit scans and track credit consumption [4].
Risks, Costs and Security
Gemini 4 Hydrogen usage is billed at provider list pricing, while each on-demand AI Scan consumes GitHub Code Security AI credits [3][4]. Scan-policy updates are atomic when changing repository and pull-request settings together: if either change is blocked, the API returns a 422 error and applies neither. Automation should detect that failure rather than assume protection is active [4]. Restrict model availability and scan administration to approved policies and roles, and confirm GHES limitations before committing to a deployment design [2][3][4].
Where Kimbodo Comes In
Kimbodo builds and operates this in production for businesses — see our AI Application Development practice, or Estimate My AI Application.
