By GitHub is running its third Copilot model retirement wave in under two months. On October 19, six more models go dark across all Copilot surfaces — Chat, inline edits, agent mode, and code completions. This batch hits the GPT-5.x family directly and includes a Grok model added barely two months ago. If your team has explicit model dependencies in any config file, you have 27 days to fix them.
What’s Leaving on October 19
GitHub published the deprecation list on September 18. Six models going away, and their official replacements:
| Retiring Model | Replacement | Minimum Plan |
|---|---|---|
| GPT-5.5 | GPT-5.6 Sol | Pro+, Max, Business, Enterprise |
| GPT-5.4 | GPT-5.6 Sol | Pro+, Max, Business, Enterprise |
| GPT-5.4 mini | GPT-5.6 Luna | Pro and above |
| GPT-5 mini | GPT-5.6 Luna | Pro and above |
| Grok 4.5 | Grok 4.6 | Pro and above |
| Gemini 3.7 Flash | Gemini 3.8 Flash | Pro and above |
Most of these are clean swaps with better successors already available. The Gemini and Grok upgrades are unambiguous improvements. The GPT-5.x replacements require a decision.
Sol vs Luna: The Decision You Actually Have to Make
The migration for GPT-5.5 and GPT-5.4 is not as simple as “use the new model.” GitHub’s replacement for the higher-tier GPT-5.x models is GPT-5.6 Sol — available only on Pro+, Max, Business, and Enterprise. If your team is on a base Pro plan and was using GPT-5.5, Sol is not accessible to you. The fallback is GPT-5.6 Terra or GPT-5.6 Luna.
For the mini variants (GPT-5.4 mini and GPT-5 mini), GPT-5.6 Luna is the clear match. Luna is the lightweight tier of the 5.6 family — built for speed and cost efficiency, available across all Pro plans. At roughly 20x lower cost than Sol, teams that chose the mini models for responsiveness or cost control should land on Luna without hesitation.
If you needed GPT-5.5’s quality ceiling and you’re on Business or Enterprise, Sol is the right move. It handles large-codebase analysis and long-running agent work better than any other model in the family. Getting the tier wrong — routing GPT-5.5 workloads to Luna to save cost — will be noticeable in output quality.
Grok 4.5 Is Already Being Retired
Grok 4.5 was added to Copilot on July 28. It’s being retired October 19. That’s 83 days. The model barely had time to appear in production configs before it’s getting removed. Its successor, Grok 4.6, is a genuine upgrade — better sustained reasoning on long-horizon tasks — but the churn rate is worth noting. Any team that spent time evaluating and adopting Grok 4.5 is now doing that work again.
What Admins Need to Check
If you’re on Copilot Business or Enterprise, the replacements are likely already enabled through GitHub’s global default model policy (GA since August 26). Under that policy, newly available GA models are on by default. You may already be set.
If your organization has disabled the global default or explicitly blocked specific models, go to your enterprise’s AI controls tab and manually enable Sol, Luna, Grok 4.6, or Gemini 3.8 Flash before October 19.
Three things to audit now:
- VS Code settings.json — Check for
github.copilot.chat.modelsentries with explicit model names - JetBrains plugin configs — Any Copilot model preferences set at the IDE level
- CI/CD integrations — Scripts or workflows calling the Copilot API with a hardcoded
modelparameter
The Pattern Is the Point
This is the third Copilot model retirement wave since September 1. Six on September 1, four on October 2 (still upcoming), and six more on October 19. GitHub is treating model versions like patch releases, not like infrastructure your team builds around.
The practical takeaway: stop pinning specific Copilot model versions in production configs. Where GitHub allows it, use plan-level defaults or tier-based selection rather than naming a specific model. The models that exist today may not exist in 90 days.
