Microsoft MAI Code 2: a shopkeeper eyes a brass balance scale that holds one stone level with a heap of stones (AI image)

Illustration

New tool3 min read

Microsoft MAI Code 2: the 5% figure runs on a promo price

Microsoft MAI Code 2 is Microsoft's new coding model, launched at Ignite on November 17, 2026 and available now in GitHub Copilot. Microsoft says it delivers 96% of Claude Opus 5.5's performance at 5% of the cost – measured on Microsoft's own eval platform, with a 50% promotional price.

Key facts

  1. Microsoft launched MAI Code 2 and MAI Code 2 Flash at Ignite on November 17, 2026; both are available in GitHub Copilot, and Code 2 Flash is also in limited preview in Microsoft Foundry.
  2. Microsoft says MAI Code 2 reaches, averaged over four benchmarks, 96% of Claude Opus 5.5's performance at 5% of the cost and 98% of GPT-6.1 Sol's at 40% of the cost.
  3. The tests ran on Microsoft's internal MSBench platform and MAI Code 2's cost uses a 50% promotional price; Microsoft gives no per-million-token price.

Manfred „Luigi“ LugmayrBy Manfred „Luigi“ Lugmayr

Researched and written with AI, to Luigi's brief, positions and editorial rules. Automatically checked against the sources. How I work →

96% for 5%. That number tops nearly every Ignite headline tonight. It's even true – inside the frame Microsoft drew for itself. And the frame is the actual story.

What Microsoft MAI Code 2 does, per Microsoft

Microsoft AI used Ignite to ship two coding models: MAI Code 2 for production code, deep debugging, multi-file changes and UI work, and MAI Code 2 Flash for high-volume routine – explaining code, small fixes, targeted edits. Both are in GitHub Copilot; Code 2 Flash is also in limited-access preview in Microsoft Foundry. OpenRouter, Vercel and others are "coming soon," no dates.

The vendor numbers, averaged over four benchmarks:

  • MAI Code 2 vs. Claude Opus 5.5: 96% of the performance at 5% of the cost.
  • MAI Code 2 vs. GPT-6.1 Sol: 98% at 40% of the cost.
  • Code 2 Flash: comparable to Claude Haiku 5.5 at 20% of the cost.

Sounds like a price war. It is. Now read the fine print on the same page.

The catch: Microsoft is referee and player

The runs happened on MSBench, Microsoft's internal eval platform, through the GitHub Copilot CLI, no subagents, every model at max reasoning effort. MAI Code 2's cost is calculated at a 50% promotional price. There's no per-million-token price on the page. And Microsoft itself says the model isn't tuned yet for very long-horizon agentic work – exactly where Opus earns its premium.

So the 5% isn't a benchmark. It's a price tag with an expiry date.

The second-order point, for me, is elsewhere. With the predecessor, MAI-Code-1.1-Flash, Copilot Free users only got the model through auto model selection. When one company owns the platform, the model and the "Auto" button, model choice is also a margin decision. That's a value transfer: you save credits, Microsoft saves on what it pays OpenAI and Anthropic, and who decides which model writes your code shows up in no dashboard. Meanwhile Microsoft says a sibling, MAI Work Flash, already serves Microsoft 365 products, in Word already in production, per Microsoft. After Microsoft's Copilot cost model from Ignite and Copilotforce with Salesforce, this is the third piece of one strategy: workspace, context, and now the model, all from one vendor.

What I'd do now

  • Open your Copilot policies. For the predecessor, the MAI policy was off by default for Business and Enterprise. Decide who gets which model instead of letting "Auto" run.
  • Your own test set beats the vendor chart: five real tasks from your repo, landing-page fix to tracking script, and measure cost per usable result. More on that in cost per task, not token price.
  • Re-run the math when the promo ends. Without the 50% discount, Microsoft's own numbers put it at roughly 10% of Opus cost – still cheap, different headline. And the blocker question from GPT-6.2 Astra applies here too: who tells you when something is missing?

On approvals for agents, see the guide AI agents for marketing.

Bottom lineWhoever owns the platform, the model and the Auto button grades the test – and sets the price while they're at it.

  • Share

Got AI news? Submit news →

FAQ

What is Microsoft MAI Code 2?

A coding model from Microsoft AI, launched at Ignite on November 17, 2026, for production code, debugging, multi-file changes and UI work. It runs in GitHub Copilot.

Is MAI Code 2 really as good as Claude Opus 5.5?

Microsoft says it reaches 96% of the performance averaged over four benchmarks, measured on its own MSBench platform. Microsoft says it isn't tuned yet for very long agentic tasks.

How much does MAI Code 2 cost?

Microsoft gives no per-million-token price. The 5%-of-Opus figure is calculated at a 50% promotional price.

Source

Microsoft AI: Introducing MAI Code 2 and MAI Code 2 Flash

More sources

  • GitHub Changelog – Vorgänger-Modell, 11.08.2026: Auto-Modellwahl, Admin-Richtlinie standardmäßig aus, Abrechnungsfaktor.

Black Friday 2026

Black Friday deals on AI tools, straight to your inbox.

I collect the AI tool deals worth having and send them in the AI Briefing, along with the AI news that matters, every Wednesday and Friday. Five minutes to read.

  • Black Friday deals on AI tools, collected for you
  • The key AI news, briefly explained
  • Wednesday and Friday, unsubscribe in one click

In the AI Briefing right now

Image for the story: Hootsuite's sale: the 50% ends a week before Black FridayDealHootsuite's sale: the 50% ends a week before Black Friday
Image for the story: Java4Shell: AI found the Java hole in an hour for under $200NewsJava4Shell: AI found the Java hole in an hour for under $200
Image for the story: Suno Visuals turns songs into videos and doubles the rights questionNewsSuno Visuals turns songs into videos and doubles the rights question
Blonde woman with glasses smiling as she scrolls on her phone in an office

Manfred “Luigi” LugmayrFrom Manfred “Luigi” Lugmayr: 30 years in marketing, seven of them as CMO at burgerme.