GitHub Copilot Drops Six Models on Oct 19: What Replaces GPT-5.5, GPT-5.4 and Grok 4.5

GitHub will pull six models out of Copilot on 19 October 2026: GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Gemini 3.7 Flash and Grok 4.5. GitHub announced the cut in a changelog post on 18 September and asked customers to "update your workflows and integrations to use the supported models before this date."

Each retiring model has a named successor. For most Copilot Business and Enterprise organizations the successors switch on by themselves. That default does not apply everywhere, though. If an admin turned off the global default, or explicitly disabled a successor, developers lose the old model and get nothing in its place. Individual Pro subscribers have a different gap: GitHub's suggested successor for GPT-5.4 is not on their plan at all.

The six models and their replacements

The OpenAI side folds four models into two. Both full-size models, GPT-5.5 and GPT-5.4, map to GPT-5.6 Sol, the top of the GPT-5.6 family. Both small models, GPT-5.4 mini and GPT-5 mini, map to GPT-5.6 Luna, the lightweight one. Google and xAI each swap one Flash or Grok version for the next.

GitHub says no clean-up is needed afterwards: "No action is required to remove the models after they have been deprecated."

Who gets the replacement automatically

Copilot Business and Enterprise have a policy called Default availability for released models. It decides whether a GA model that an admin has never configured is on or off. Models in that state carry a Delegate to Default Policy label in the enterprise or organization model settings.

The deprecation notice spells out how that policy applies here: "Under default model enablement, the suggested alternatives are automatically enabled for Copilot Enterprise and Copilot Business customers unless an administrator has turned off the global default or explicitly disabled the model."

What developers see on 19 Oct depends on the Default availability for released models policy and on whether the successor was disabled.

That leaves three cases for an organization:

1. Global default on, successor never configured. GPT-5.6 Sol, GPT-5.6 Luna, Gemini 3.8 Flash and Grok 4.6 are already on, or come on without anyone touching them. Developers who pinned an old model just need to pick the new one.
2. Global default off. The successors stay off. When the old models disappear on 19 October, the org has no route to their nearest replacements. GitHub's instruction is to "enable access to the alternative models through their model policies in Copilot settings."
3. Successor explicitly disabled. The same result as case 2, but for that one model. A team that blocked GPT-5.6 Sol on purpose, perhaps during an evaluation, has to make a call now. It can unblock Sol or move its GPT-5.5 and GPT-5.4 users onto something else.

When GPT-5.6 arrived in Copilot on 9 July, GitHub's launch note told Business and Enterprise admins that the GPT-5.6 policy "is off by default" and had to be enabled in Copilot settings. Any org that read that note and left the policy off, or set it to disabled while it evaluated the models, is in case 2 or case 3. It should check before 19 October.

The Pro gap: GPT-5.4 users have no Sol

GitHub's per-plan model table lists GPT-5.4 as included on Copilot Pro. GPT-5.6 Sol is not: Sol is available on Pro+, Max, Business and Enterprise only. GPT-5.5 is not on Pro either.

So a Pro subscriber who uses GPT-5.4 loses it on 19 October and cannot switch to the suggested alternative without upgrading. The GPT-5.6 models Pro does include are GPT-5.6 Terra and GPT-5.6 Luna. GitHub's pricing page puts GPT-5.4 and Terra in the same "Versatile" class, which makes Terra the closest in-plan substitute. That is our reading of the tables; GitHub's notice does not name Terra.

The small models are not affected this way. GPT-5.4 mini and GPT-5 mini users on Pro go to GPT-5.6 Luna, which is on every paid plan.

What the switch costs

Copilot now converts every model's token usage into GitHub AI Credits, at $0.01 a credit, so a model swap changes what each task costs. (Pro and Pro+ subscribers still on an older annual plan with request-based billing use model multipliers instead.) Here are GitHub's list rates per 1M tokens in the default context tier:

Output price per 1M tokens in the default context tier, from GitHub's Copilot pricing table. The two panels use different scales.

Most of the moves are cheaper or cost the same:

• GPT-5.5 → Sol cuts output from $30.00 to $20.00 per 1M tokens and input from $5.00 to $4.00.
• Both mini models → Luna get cheaper. GPT-5.4 mini's $4.50 output rate drops to $1.20.
• Gemini and Grok are price-for-price swaps. Gemini 3.7 and 3.8 Flash are both on promotional pricing of $0.75 input and $3.75 output through 31 December 2026.
• GPT-5.4 → Sol is the one move that costs more. Output goes from $15.00 to $20.00 and input from $2.50 to $4.00. A team that chose GPT-5.4 because it was cheaper than GPT-5.5 is being moved up a price class. GPT-5.6 Terra, at $2.00 input and $12.00 output, is cheaper than the model it replaces.

The GPT-5.6 models also bill a cache-write charge that earlier OpenAI models in Copilot did not ($5.00 per 1M tokens for Sol, $0.25 for Luna). Long agent sessions that lean on prompt caching should be checked against real usage, not the headline rates alone.

What to do before 19 October

Business and Enterprise admins

• Open the enterprise or organization model settings and check the Default availability for released models policy.
• If it is off, enable GPT-5.6 Sol, GPT-5.6 Luna, Gemini 3.8 Flash and Grok 4.6 individually in model policies. Add GPT-5.6 Terra if GPT-5.4 users would be better served by a cheaper model.
• If it is on, look for any of those four that someone explicitly disabled. A disabled successor is not switched on by the default.
• Tell developers which model replaces the one they pinned. Check budgets and spend caps for teams moving from GPT-5.4 to Sol.

Pro subscribers

• If you use GPT-5.4, try GPT-5.6 Terra now, or budget for Pro+ if you need Sol specifically.
• If you use GPT-5.4 mini or GPT-5 mini, switch to GPT-5.6 Luna.

Everyone with integrations

• Search scripts, Copilot CLI configs and agent definitions for hard-coded model names (gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5-mini, Gemini 3.7 Flash, Grok 4.5). GitHub asks for "workflows and integrations" to be updated before the date. A pinned name that no longer exists will fail.
• Check the minimum client versions. GitHub's support table lists VS Code 1.128.0 for the GPT-5.6 models.

Three days later another default flips. GitHub's documentation says a separate Default policy for new features starts applying on 22 October 2026, and unconfigured GA Copilot features will be enabled unless an admin acts. Admins already in the policy settings for this model change can review that one in the same pass.

Sources

• GitHub Changelog: Upcoming deprecation of selected GitHub Copilot models in mid-October (18 Sep 2026)
• GitHub Changelog: OpenAI's GPT-5.6 Sol, Terra, and Luna are now available in GitHub Copilot (9 Jul 2026)
• GitHub Docs: Supported AI models in GitHub Copilot
• GitHub Docs: Models and pricing for GitHub Copilot
• GitHub Docs: About default availability of Copilot features and models