Skip to content

MAI Code 2 and MAI Code 2 Flash arrive in GitHub Copilot

GitHub added two Microsoft AI coding models to Copilot's model picker, one built for heavy agentic work and one for cheap everyday edits.

Draftship Editorial3 min read

On November 17, 2026, GitHub added two Microsoft AI coding models to GitHub Copilot: MAI Code 2 and MAI Code 2 Flash. Both appear in the model picker, and the changelog entry announcing them describes the pair as a step forward for the MAI-Code family, which it says is expanding from lightweight coding assistance to complex, agentic software engineering. MAI Code 2 in GitHub Copilot is the heavier of the two; Flash is the lighter one, and the only one of the pair that reaches the Free and Student plans.

What each model is meant to do

GitHub splits the two by task weight rather than by feature set.

MAI Code 2 is described as a general purpose coding model built to return focused responses and practical code changes with less unnecessary back and forth. GitHub positions it for deep debugging, multi-file implementation, refactoring and reasoning across a whole repository.

MAI Code 2 Flash is the lightweight, cost-efficient option aimed at everyday work: explaining code, making targeted edits, routine fixes and fast iteration.

Both are described as designed for agentic coding workflows, so the choice is about matching the model to the task in front of you rather than settling on a single default.

Where you can pick them, and who gets access

MAI Code 2 is listed for Copilot Pro, Pro+, Max, Business and Enterprise. Flash goes wider: Copilot Free, Student, Pro, Pro+, Max, Business and Enterprise. GitHub says both are available to eligible Copilot subscribers on annual plans.

Billing is usage-based. The changelog points to GitHub's models and pricing documentation for the numbers and does not include any rates itself.

The picker appears in Visual Studio Code, Visual Studio, Copilot CLI, the GitHub Copilot cloud agent, the GitHub Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains, Xcode and Eclipse.

For Business and Enterprise customers, administrators manage access through the model policy in Copilot settings. Under default model enablement, new models are turned on automatically unless an administrator has switched off the global default or explicitly disabled that model. Teams that keep a tight allowlist should check the policy rather than assume nothing changed.

What likely changes for people building apps

Two tiers under one billing scheme is mostly a cost-control question, and this suggests routing becomes the point. Routine requests, such as explaining a stack trace or renaming a variable, are the case GitHub assigns to Flash. Refactoring across several files or tracing a bug through a repository is the case it assigns to MAI Code 2. Both are available on the same paid plans, so the decision is which model to attach to which request.

The Free and Student tiers gaining Flash is likely the bigger shift for people without a paid seat: a model GitHub describes as optimized for code explanations and rapid iteration is now reachable on those plans, which likely lowers the cost of learning or prototyping.

Limits and open questions

The changelog is thin on hard numbers. It gives no benchmark scores, no context window sizes and no latency figures, and it defers per-model pricing to separate documentation. It also points readers to a separate MAI Code 2 announcement for detail on vision support, coding performance improvements and efficiency gains, none of which appears here.

It is unclear whether usage rates differ between the two models, and the annual-plan wording is not explained for subscribers billed another way.

What this means for building with AI

Model choice inside a coding assistant is turning into a routing decision: cheap and fast for the routine work, heavier for the reasoning. Anyone building an app with an AI tool now has to think about which model sits behind which request, not just whether AI is involved. If you are working in Draftship, AI chat and model settings explain how model selection works on paid plans. Draftship is not affiliated with GitHub or Microsoft, and nothing here suggests Draftship supports these models.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.

AI coding agents

opencode v2.3.2 fixes GPT-6.2 prompt caching billing

opencode v2.3.2 sends the system prompt as a leading developer message on every Responses route so GPT-6.2 caches it, cutting uncached input per step from about 14,000 tokens to 3 on one Copilot model.