Microsoft's MAI Code 2: cheaper coding models, already in GitHub Copilot
Microsoft released MAI Code 2 and Code 2 Flash at Ignite on 17 November. They are live in GitHub Copilot, with claims of near-frontier coding at a fraction of the cost. What to check before you switch.
- Author
Umair Mateen- Published
- Reading time
- 4 min
- Category
- AI
Ideas in this post
Tap one for a 5-second answer. Underlined words in the article work the same way.
0 of 4 understood

Microsoft released MAI Code 2 and MAI Code 2 Flash on 17 November 2026, at its Ignite conference: two coding models built by its own Microsoft AI (MAI) team rather than licensed from OpenAI or Anthropic. Both are already selectable in GitHub Copilot, and Microsoft's pitch is simple: near-frontier coding quality for a fraction of the price.
What Microsoft released
- MAI Code 2 is aimed at everyday enterprise coding: building features, fixing bugs, work.
- MAI Code 2 Flash is the smaller, cheaper model for high-volume, routine jobs such as explaining code, small targeted edits and simple fixes.
Both are part of a wider batch of new MAI models announced the same day, alongside a cyber-defence model, image models, a live model and a routing model (a that decides which bigger model should handle a request), listed on Microsoft's Ignite models page.
How Microsoft says they compare
Microsoft's own figures, averaged across four coding (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1):
Quick check · Everyday ↔ real
The restaurant and your app, side by side
Switch between the two and see how each part lines up.
- Dining room and menu
- Placing your order
- The kitchen door
- The kitchen
- The pantry
Guests see only the dining room. The kitchen does the work out of sight, and the dish comes back to the table.
| Everyday | Real |
|---|---|
| Dining room and menu | Frontend: screens and buttons |
| Placing your order | Clicking "Buy" |
| The kitchen door | A request over HTTP |
| The kitchen | Backend: servers and rules |
| The pantry | The database |
From the explainer Frontend vs backend, explained with a restaurant
- MAI Code 2 reaches 98% of the score of OpenAI's GPT-6.1 Sol at 40% of the cost, and 96% of Claude Opus 5.5's score at 5% of the cost.
- MAI Code 2 Flash is described as comparable to GPT-6 Luna and Claude Haiku 5.5, at 65% and 20% of their cost respectively.
- In blind front-end comparisons with 258 reviewers, people preferred MAI Code 2's output over GPT-6 Sol's about two times in three.
Read these with care. They are vendor benchmarks, run by Microsoft at maximum reasoning effort inside GitHub Copilot CLI. The footnotes also say the MAI costs in the comparison are shown at a 50% promotional price, so the gap may shrink when the promotion ends. Microsoft itself says the models are not yet tuned for scientific research or very long-running agent work.
The model page also carries two customer stories: AT&T says a custom version of Code 2 Flash beat GPT-6.1 Sol on its 5G network-analysis tasks at far lower cost, and Woven by Toyota says a fine-tuned Code 2 Flash produced ready-to-merge fixes in 44% of cases where frontier models had needed human rework.
Where you can use it
- GitHub Copilot: per GitHub's changelog, MAI Code 2 is available on Pro, Pro+, Max, Business and Enterprise plans (not Free or Student); Code 2 Flash is on every plan including Free and Student. Both use usage-based billing, and the changelog refers to eligible subscribers on annual plans, so check your own plan.
- Editors: the model picker in VS Code, Visual Studio, JetBrains, Xcode, Eclipse, Copilot CLI, the Copilot cloud agent, github.com and GitHub Mobile.
- Microsoft Foundry: Code 2 Flash is in limited-access preview for developers who want to call it from their own apps.
- Elsewhere: Microsoft says both are coming soon to OpenRouter, Vercel, Hermes, Open Code and Kilo. No date given.
Microsoft has not published per-token API prices, a context window size or rate limits in these announcements.
Who should care
If your team pays for GitHub Copilot Business or Enterprise, this matters now: GitHub says new models are switched on automatically under default model settings unless an admin turns that off. That means developers may already see MAI Code 2 in their picker today, and their usage will be billed by consumption.
For companies spending heavily on premium coding models, a cheaper model that handles most routine work could cut costs noticeably, with the expensive model kept for hard problems. That is the same pattern behind model routing generally: cheap model by default, premium model when needed.
What to do this week
- Admins: check your Copilot model policy and decide whether MAI Code 2 and Code 2 Flash should be enabled, and for whom.
- Run your own test: give the same five or ten real tickets to your current model and to MAI Code 2, and compare review time, not just whether the code runs.
- Watch the bill: note that the headline savings use a promotional price, and set spending alerts on usage-based billing.
- Keep the big model for big jobs: Microsoft itself steers long agent runs and research tasks elsewhere for now.
The DevDrift helps teams set up AI coding tools with sensible model choices, reviews and cost limits.
Sources
- Microsoft AI: MAI Code 2 and Code 2 Flash: Everyday coding at scalable prices (17 November 2026)
- GitHub Changelog: MAI Code 2 and MAI Code 2 Flash in GitHub Copilot (17 November 2026)
- Microsoft AI: Seven new world class models at scalable price-performance (Ignite 2026)
- Microsoft AI: MAI Code 2 and Flash model page
You’ve read the news. Now understand it.
- Learn · 5 min · Everyday ↔ realFrontend vs backend, explained with a restaurantThe frontend is the part of an app people see and use; the backend is the hidden part that stores data and applies the rules. Here's the difference, explained with a restaurant's dining room and kitchen.Start in 5 seconds →
- Learn · 5 min · Side by sideWhat is a voice agent? Explained with a receptionist who never puts you on holdA voice agent is an AI you talk to out loud that listens, checks your systems and answers in a spoken voice. Explained with a front-desk receptionist who answers every call straight away.Start in 5 seconds →
- Learn · 5 min · Side by sideSmall vs large AI models, explained with a bicycle courier and a lorryLarge frontier AI models handle the hardest work but cost more and run slower; small models are quick and cheap for focused jobs. Explained with a bicycle courier and a lorry.Start in 5 seconds →
- Learn · 5 min · Myth or factWhat are AI benchmarks? Explained with a school examAn AI benchmark is a standard test: the same questions and the same marking for every model, so scores can be compared. Explained with a school exam that every student sits.Start in 5 seconds →


