GitHub put Microsoft's MAI-Code-1.1-Flash into Copilot on 11 August and, in the same breath, gave its predecessor a retirement date. That pairing is the whole story: this is a swap, not an addition.
The numbers are the interesting part. A 0.25× premium request multiplier for annual subscribers and a list price 73% below the model it replaces make this the cheapest way to run a lot of routine work through Copilot.
What it is
MAI-Code-1.1-Flash is Microsoft's latest small-tier coding model. GitHub lists three improvements over the previous version: native vision support for understanding images, better coding quality and instruction following, and better tool use.
Native vision in a small, cheap model is more useful than it sounds. It is what lets you paste a screenshot of a broken layout or a stack trace image into chat and have the agent work from it, and until now that meant reaching for a much more expensive model.
Who gets it, and who has to switch it on
The rollout differs by plan in a way worth reading twice:
- Free and Student — auto-selected, so it becomes the model you get without choosing
- Pro, Pro+, Max, Business and Enterprise — manually selectable
- Business and Enterprise — off by default; an administrator has to enable the policy first
Available almost everywhere
The surface list is close to complete: the Copilot CLI, the cloud agent, GitHub's desktop app, chat on github.com, Visual Studio Code, Visual Studio, GitHub Mobile, JetBrains IDEs, Eclipse and Xcode.
That breadth matters more for a cheap model than for an expensive one. A small model earns its place by being the default for volume work — the quick edits, the boilerplate, the routine questions — and it can only do that if it is available everywhere the volume happens.
The 10 September deadline
MAI-Code-1-Flash is deprecated on 10 September 2026. GitHub's instruction is direct: update your workflows and integrations to use the new model before that date.
No action is required to remove the old model once it is deprecated — it simply becomes unavailable. That is the part that bites. If something automated is pinned to the old model name, it does not degrade gracefully on 10 September; it stops.
Enterprise administrators have a short checklist: enable MAI-Code-1.1-Flash in model policies, confirm the policy is on, and check that developers can actually see it in the model selector. Doing that in August rather than on 10 September is the difference between a change and an incident.
What it means for your bill
Two different pricing statements apply, and mixing them up is easy. There is a 0.25× premium request multiplier for annual subscribers, and separately a list price 73% below MAI-Code-1-Flash for usage billed against provider pricing.
The practical effect is that routing routine work to the small model is now clearly the cheapest option in Copilot, and the case for making an expensive model your default is weaker than it was a week ago. The right pattern is not "always use the cheap model" but "use the cheap model until it fails", which requires you to actually notice when it fails — which requires reviewing the output rather than accepting it.