What shipped
GitHub's changelog splits the two models by job:
- MAI Code 2 is the general-purpose model. GitHub points it at deep debugging, multi-file implementation, refactoring and repository-wide reasoning.
- MAI Code 2 Flash is the lightweight, cost-efficient option. GitHub describes it as suited to code explanations, targeted edits, routine fixes and rapid iteration.
Both are described as designed for agentic coding workflows. GitHub frames this as letting developers choose the model that fits the task.
Who gets it, and where
Availability depends on your plan:
- MAI Code 2: Copilot Pro, Pro+, Max, Business and Enterprise.
- MAI Code 2 Flash: Copilot Free, Student, Pro, Pro+, Max, Business and Enterprise.
GitHub adds that both models are available to eligible Copilot subscribers on annual plans. The changelog doesn't say what "eligible" covers, so check your own plan if the models don't appear.
The model picker lists these clients:
- Visual Studio Code and Visual Studio
- Copilot CLI
- GitHub Copilot cloud agent and the GitHub Copilot app
- github.com
- GitHub Mobile on iOS and Android
- JetBrains IDEs, Xcode and Eclipse
The changelog gives no minimum client versions. It also doesn't say whether every client behaves identically, so update your editor or CLI if the new entries are missing.
Admin controls
Business and Enterprise administrators manage access through the model policy in Copilot settings. Under default model enablement, new models switch on automatically unless an admin has turned off the global default or explicitly disabled the model.
Organizations with strict change control should check that policy now. A model arriving unannounced in developers' pickers is the kind of surprise that tends to show up in an audit. A plan being listed as eligible doesn't guarantee access if policy blocks it.
The pricing
Usage is metered. GitHub's pricing documentation says the models are billed under usage-based billing, and interactions consume input, output and cached tokens. Those tokens are converted into AI credits at 1 credit = $0.01.
The listed rates per 1 million tokens are:
| Model | Input | Cached input | Output |
|---|---|---|---|
| MAI Code 2 | $0.50 | $0.025 | $2.50 |
| MAI Code 2 Flash | $0.05 | $0.005 | $0.25 |
For context, the same table lists GPT-6.1 Sol at $2.00 input and $10.00 output in its default tier. It lists Claude Opus 5.5 at $4.00 input and $20.00 output.
These are promotional prices. The documentation footnotes say MAI Code 2 pricing runs until May 17, 2027, then doubles to $1.00 input, $0.05 cached and $5.00 output. Flash moves to $0.10, $0.01 and $0.50 on the same date. The trade site RuntimeWire made the same observation about the doubling. Microsoft's own cost comparisons also use the discounted rate, so the savings it cites will shrink once the promotion ends.
To cap spending, GitHub's documentation describes a workaround. Create or edit a Bundled AI credits budget, set it to $0, and enable "Stop usage when budget limit is reached." Business and Enterprise plans use per-user allowances pooled at the billing-entity level.
Microsoft's performance claims
Microsoft AI's companion post makes the headline claims. All of them are Microsoft's own, not independent results:
- Averaged over four benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1), MAI Code 2 reaches 98% of GPT 6.1 Sol's performance at 40% of the cost. It also reaches 96% of Claude Opus 5.5's performance at 5% of the cost.
- In blind frontend evaluations, reviewers preferred its results over GPT-6 Sol 67% of the time.
- Flash is said to match GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of the cost, across three routine-coding benchmarks.
The methodology has limits. All models ran at max reasoning effort, and three of the benchmarks ran on Microsoft's internal MSBench platform using the GitHub Copilot CLI without subagents. That setup tells you how the models do in one harness. It doesn't tell you how they behave in your monorepo with your tooling.
Microsoft also says the models are built for everyday enterprise coding and are explicitly not yet tuned for fundamental science research or very long-horizon agentic work. That caveat sits awkwardly beside GitHub's "complex, agentic software engineering" language. Treat Code 2 as a strong everyday workhorse, not a model for open-ended autonomous projects.
Coverage elsewhere has repeated claims about AT&T evaluations and a 50x cost advantage for Flash. One outlet noted that independent benchmarks have not verified the performance claims. I wouldn't plan around those figures.
Other details
- Microsoft says Code 2 Flash is also in limited-access preview in Microsoft Foundry.
- Microsoft says it will release the models on third-party platforms, including OpenRouter and Vercel, later.
- Microsoft also describes a derivative model, MAI Work Flash, for Excel, PowerPoint and Word. That is a separate topic and not part of this Copilot release.
How to evaluate it
- Check access. Update your client, open the model picker, and look for both models. Admins should review the model policy first.
- Pilot Flash on routine work. Explanations, small edits and routine fixes are where its cost advantage matters most.
- Try Code 2 on a real multi-file task. Compare the diff quality and the number of correction rounds against your current default.
- Watch the credit meter. Compare token consumption across models on the same task, and set a budget limit if needed.
- Plan for May 2027. Don't build a budget on promotional rates.
Bottom line
Microsoft's pitch is near-frontier everyday coding at a small fraction of frontier cost. The pricing is concrete, the access list is clear, and the admin controls are familiar. The benchmark claims are Microsoft's own, run on its own platform, and priced at a temporary discount. The models look worth a pilot, and independent testing should come before any broad switch of default models.
References
- MAI Code 2 and MAI Code 2 Flash in GitHub Copilot GitHub Changelog · 2026-11-17T18:01:45+00:00
- Models and pricing for GitHub Copilot - GitHub Docs docs.github.com
- MAI Code 2 and Code 2 Flash: Everyday coding at scalable prices microsoft.ai