Microsoft Adds Two MAI Coding Models to GitHub Copilot
Microsoft has added MAI Code 2 and MAI Code 2 Flash to GitHub Copilot, giving developers a choice between a larger model for complex repository work and a smaller model for routine edits. The November 17 announcement is a released-product story, with separate access conditions for each model.
Who can use the new models
GitHub lists MAI Code 2 for Copilot Pro, Pro+, Max, Business and Enterprise. Flash also reaches Free and Student users. Eligible annual subscriptions are included. Business and Enterprise administrators can control model access through policy, so a supported subscription does not by itself guarantee that an employee can select the model.
GitHub lists support across its model picker in VS Code, Visual Studio, Copilot CLI, cloud agent, app, website, mobile apps, JetBrains, Xcode and Eclipse. Developers should check their own client and organisation settings before assuming availability.
The price claim needs its footnote
Microsoft promotes MAI Code 2 using comparisons with other coding models. These figures describe selected benchmarks, reasoning settings and pricing assumptions. Its launch notes say the MAI cost shown reflects a 50 percent promotional price. That makes an unqualified claim of permanent savings misleading.
The practical question for a team is the cost of accepted work. A cheaper model call can still require extra attempts, human review or a stronger model to finish a difficult change. Conversely, a smaller model may be sufficient for a narrow edit. Those are evaluation questions, not outcomes established by this announcement.
What differs between Code 2 and Flash
Microsoft positions Code 2 for deeper debugging, changes spanning multiple files and reasoning across a repository. Flash targets explanations, focused edits and repeated small fixes. This is a task-based distinction. It does not establish that either model can safely approve its own changes or replace normal testing and review.
What the model card adds
Microsoft’s November 17 model card describes both models as accepting text and images and producing text. It lists a total context length of 1,050,000 tokens, with separate input and output limits. Those limits describe capacity, not a promise that every long request will produce a correct answer. For developers, the useful distinction is whether a task needs visual input, such as a screenshot of a broken interface, alongside repository files. Test that combination on the actual client you plan to use.
What Indian developers should check first
The launch is relevant to developers in India who already use Copilot, but it is not an India-specific hiring announcement. Microsoft has not attached a vacancy count, salary claim or employment promise to these models. Students should also distinguish access to a model from unlimited use of it. A plan can include a model while still imposing usage rules and billing conditions.
Start with the account that you actually use for work. A personal subscription and an employer-managed account can expose different model choices. If a model does not appear, check the client version, subscription eligibility and organisation policy. Buying another subscription before checking those settings could solve the wrong problem.
A useful trial needs a real repository
For a small development team, a sensible evaluation starts with a completed issue whose correct solution is already known. Give each candidate model the same repository state, request and tool permissions. Keep the test separate from customer production systems. This is an editorial recommendation for comparing tools, not a result of Microsoft’s benchmark.
Record whether the proposed change builds, passes relevant tests and addresses the actual requirement. Also record the time a reviewer spends understanding and correcting it. A patch that passes a narrow test can still introduce an unnecessary dependency or make future maintenance harder. Looking only at the first response misses those costs.
Use more than one task. An explanation request, a small bug fix and a change spanning several files exercise different abilities. Keep a record of failures, including cases where the model reports success without completing the work. That makes the comparison useful when deciding which tasks can go to Flash and which need a larger model or direct human attention.
Keep permissions proportional to the task
A coding agent may need to read files, run commands or access development services. Review those permissions before testing. Do not add production credentials merely because a task is difficult. A model’s advertised coding ability does not establish that every tool action it proposes is appropriate for your environment.
For a student project, this can be as simple as using a separate branch and checking the diff before accepting changes. For a company, existing code-review, security and deployment controls still apply. The launch does not supply evidence that those controls can be removed.
What remains unknown
The announcement does not tell an Indian team how much its own monthly bill will change. That depends on task volume, the mix of models, output length, retries and account billing. It also does not establish a universal winner for every programming language or repository. Published comparisons are a starting point for a trial, not a substitute for one.
Access through GitHub Copilot should also be kept separate from other distribution plans. Microsoft’s specialist announcement describes Code 2 Flash in Foundry as a limited-access preview and names additional platforms as coming later. A planned integration should not be reported as available simply because the model is already present in another product.
Frequently asked questions
Yes. GitHub lists both in Copilot for eligible plans. Organisation administrators can control access. Other distribution channels have separate availability conditions.
GitHub lists Flash for Free users. Its published list for the larger Code 2 model begins with Pro. Check current account limits and billing before starting a large task.
The launch comparisons use promotional MAI pricing and specified evaluation settings. They do not establish permanent savings for every workload or account.
No job cuts are established by these release documents. The verified event is the launch of coding models and their availability in development tools.
Related BoringDude coverage
For another example of model pricing and access conditions, read our coverage of Anthropic’s API price changes. The products and restrictions differ, so compare the actual workload and access terms rather than treating every advertised price cut as equivalent.
Source record
Microsoft AI launch and benchmark conditions
GitHub availability and policy details

Average Rating