GitHub is retiring six Copilot models and has rebuilt the code review overview

GitHub published two changes to Copilot on September 18, 2026. The first sets a deadline: six models are being deprecated on October 19, 2026, with a replacement named for each. The second is already live: Copilot code review reached general availability, with a rebuilt overview comment, a title on every finding, and a generated commit message for batched suggestions.

What stops working, and what takes its place

The removal covers every Copilot experience at once, which GitHub lists as chat, inline edits, ask and agent modes, and code completions. Nothing has to be cleaned up afterward, because GitHub states that no action is required to remove the models once they are deprecated.

Under default model enablement, the suggested alternatives switch on by themselves for Copilot Enterprise and Copilot Business customers, unless an administrator has turned off the global default or disabled that particular model. Where the global default is off, access to a replacement is granted through its model policy in Copilot settings, and it then appears in the chat model selector.

The mapping itself is short and explicit.

• Gemini 3.7 Flash is replaced by Gemini 3.8 Flash.
• GPT-5.5 is replaced by GPT-5.6 Sol.
• GPT-5.4 is replaced by GPT-5.6 Sol as well.
• GPT-5.4 mini and GPT-5 mini are both replaced by GPT-5.6 Luna.
• Grok 4.5 is replaced by Grok 4.6.

What has to change before the date

Someone who picks a model from a menu loses nothing, because the menu changes along with the deprecation. The cost lands wherever a model is named in writing: a workflow file, a script that reaches Copilot through an integration, a shared configuration that pins one model so a whole team gets the same answers. GitHub's instruction is to update those before the date.

This suggests the real exposure is not the choice of model but the places where that choice was written down and then forgotten. GitHub does not put it that way, and enterprise customers with questions are pointed at their account team.

The code review overview was rebuilt

Copilot code review now sorts its findings into three labeled groups inside the overview comment. Open holds issues that have not been addressed, marked as new when a fresh commit introduced them. Resolved since last review holds issues Copilot has verified as fixed. Previously missed holds issues it did not raise before and found on a later pass, printed in full because they appear nowhere else on the pull request. Each finding shows its severity and links to the matching inline comment.

Every review comment now carries a short title describing the finding, and those titles are what the overview lists. Auto-resolution changed twice over: a reply asking to keep an issue open is honored, and a comment Copilot closes on its own now records a reason, either Won't Fix or Incorrect. The overview preserves that record as further commits arrive, alongside the earlier pull request and per-file summaries.

Committing a complete batch of eligible suggestions now produces a generated commit title and an optional description drawn from the selected changes, and a batch may include comments that did not come from Copilot. GitHub says these updates are generally available.

What actually changes in the review

Two of the four changes are about the record rather than the finding. Previously missed exists because a reviewer that runs again over the same code can surface something it walked past the first time, and until now that arrived as an ordinary comment with nothing to mark it as late. One possible reading is that the goal was less to find more and more to make the second pass legible. The resolution reason works the same way: Won't Fix and Incorrect are two different verdicts on the same closed comment, and they used to look identical in the thread.