GitHub’s changelog entry from September 18 sets October 19, 2026 as the date six more models disappear from Copilot. The removal covers Copilot Chat, inline edits, ask and agent modes, and code completions. It is Monday. If anything in your org pins one of these models, you have roughly one working weekend left.
The list#
Straight from GitHub’s changelog:
| Retiring | Suggested replacement |
|---|---|
| Gemini 3.7 Flash | Gemini 3.8 Flash |
| GPT-5.5 | GPT-5.6 Sol |
| GPT-5.4 | GPT-5.6 Sol |
| GPT-5.4 mini | GPT-5.6 Luna |
| GPT-5 mini | GPT-5.6 Luna |
| Grok 4.5 | Grok 4.6 |
Three vendors, four product families, zero Anthropic models. The September 1 wave was mostly Claude; this one is mostly OpenAI.
What happens to your org#
For Copilot Business and Enterprise, GitHub says the replacements switch on automatically under default model enablement. That does not apply if an administrator turned off the global default or explicitly disabled a model. In that case an admin has to enable the replacements through model policies in Copilot settings before they appear in the Chat model selector. GitHub also says no action is needed to remove the retired models.
The practical risk is quiet: an org that disabled models globally, long ago, for compliance reasons may find on Tuesday that developers have a model picker with nothing useful in it.
What this does not mean#
One third-party analysis makes a point worth repeating: this is a change to model availability inside Copilot. It does not announce that OpenAI’s, Google’s or xAI’s own APIs are shutting down these model names. Do not rip out direct API integrations because of this changelog entry. I only verified that point against that analysis, not against the vendors’ own deprecation pages.
I also saw a secondary source claim an earlier October 2 wave (Gemini 3.5 and 3.6 Flash, Kimi K2.7 Code, Claude Opus 4.7). I could not find that in the official changelog results I pulled, so I am not relying on it.
The model-zoo tax, again#
I wrote about this in August and the pattern has not changed: Copilot sells breadth, and breadth means a rolling retirement calendar. Every few weeks a batch of models goes away, and each one is a pin, a policy, or a prompt tuned to a specific model’s quirks.
Replacing GPT-5.5 with GPT-5.6 Sol is not a rename. Different model, different behavior on your prompts, possibly different cost. Auto-enablement protects availability; it does not protect output quality. The tools I trust most for autonomous work are the ones where I choose the model deliberately and change it on my schedule, not a vendor’s.
That is the spec-driven angle. If your workflow depends on “whatever model the picker had last quarter”, your specs are leaking. A spec should state the outcome and the acceptance checks; the model is an implementation detail you can swap. Teams with real acceptance tests treat a forced model change as a re-run, not a crisis. Teams that validated by eyeballing diffs have to start over.
What to do before Monday#
- Grep for the names. Search repos, CI configs, agent definitions and org policy docs for
gpt-5.5,gpt-5.4,gpt-5-mini,gemini-3.7-flashandgrok-4.5. - Check your admin settings. If global default enablement is off, enable the replacement models you actually want before the cutoff.
- Re-run your acceptance suite against the replacement models, especially the mini-to-Luna swaps that likely sit behind cheap background tasks.
- Check your billing assumptions. I did not verify per-model pricing for the replacements; look at the multiplier or rate for each before it becomes the default.
- Separate Copilot from direct API use. Decide per integration whether it is affected.
Verdict#
None of this is scandalous. Retiring old models is normal and GitHub gave a month’s notice. But the cadence is the cost of the multi-model-picker approach, and it is a good moment to ask whether your team’s quality bar lives in your specs and tests or in a dropdown you do not control.
Sources: GitHub changelog, “Upcoming deprecation of selected GitHub Copilot models in mid-October” (primary, fetched directly for the model list, replacements and enablement behavior). Secondary: C# Corner, “GitHub is retiring some Copilot models” for the API-scope caveat, via search summary only.