On November 17, 2026, Microsoft launched seven in-house MAI models at Ignite 2026, and two coding models — MAI Code 2 and MAI Code 2 Flash — went live in GitHub Copilot the same day. This article covers the launch itself: what the seven models are, where MAI Code 2 fits, where it appeared on day one, and the limits Microsoft drew itself. Benchmarks and pricing get their own articles; this one establishes what was actually released.
| Item | Confirmed |
|---|---|
| Release date | Nov 17, 2026 (Tuesday), Ignite 2026 (San Francisco, Nov 17–20) |
| Released by | Microsoft AI (Microsoft), the team led by CEO Mustafa Suleyman |
| The seven MAI models | MAI Code 2, MAI Code 2 Flash (coding); MAI Cyber 2 Flash (security); MAI Image 2.7 (image editing); MAI Realtime 1 (speech-to-speech); MAI Transcribe 2 (transcription, powering Dragon Copilot Transcribe); Decision 1.1 (task routing) |
| Chip | All seven models run on Microsoft's second-generation Maia 200 chip, now in production |
| MAI Code 2 positioning | Everyday enterprise coding flagship: production code, deep debugging, multi-file changes, repo-wide reasoning, UI work; also tuned for dashboards and visualizations, per Microsoft |
| MAI Code 2 Flash positioning | Lightweight tier: code explanations, targeted edits, routine fixes, rapid iteration |
| Where it launched | GitHub Copilot (launch day): VS Code, Visual Studio, Copilot CLI, cloud agent, GitHub.com, mobile, JetBrains, Xcode, Eclipse; Flash also in limited-access preview in Microsoft Foundry |
| Plans | MAI Code 2: Pro, Pro+, Max, Business, Enterprise; Flash: those plans plus Free and Student |
| Billing | Usage-based inside GitHub Copilot; MAI Code 2 $0.50/$2.50 per 1M tokens (promo to May 17, 2027, then $1.00/$5.00), Flash $0.05/$0.25 (then $0.10/$0.50) |
| Weights | Closed, no public weights |
| Parameters / context / knowledge cutoff | Published in the model card: about 80B active (~2.8T total), 1.05M context, knowledge cutoff before August 2026 data |
This launch was a family debut, not a single model. Microsoft AI's Ignite post packages all seven under "price-performance": beyond the two coding models, MAI Cyber 2 Flash covers security (Microsoft-reported CyberGymE2E 92.3%, under 2% of GPT-6 Astra's cost per task), MAI Image 2.7 covers image editing (Microsoft-reported #1 on the AA image-editing board at a fifth of GPT Image 2.6 Sunburst's price), MAI Realtime 1 covers speech-to-speech, MAI Transcribe 2 covers transcription (10M Dragon Copilot patient encounters a month, error rate halved — all Microsoft-reported), and Decision 1.1 covers task routing. The other five models' figures are likewise all Microsoft-reported; this topic mentions them in passing and builds no separate topics for them — apart from Image 2.7 and Cyber 2 Flash, the material so far amounts to a single sentence each.
First, it is Microsoft's own model. Until now Copilot's workhorses were OpenAI models; MAI Code 2 is the first time Microsoft AI has put an in-house coding flagship inside its own core developer product. Copilot also added OpenAI's GPT-6.2 Astra and Sol the same day — Microsoft now offers both in-house and OpenAI options side by side in its own product.
Second, it was usable on launch day. No waitlist, no staged rollout: GitHub's changelog confirms both models entered the model picker on launch day, across nine Copilot entry points. MAI Code 2 also joined Copilot Fusion Standard the same day, so Fusion users may run into it too.
Third, the pitch is price-performance, not the capability ceiling. Microsoft reports MAI Code 2 at 98% of GPT-6.1 Sol at about 40% of the cost (Microsoft-chosen benchmarks, 50% promo-price basis), and 96% of Claude Opus 5.5 at about 5% of the cost (same basis). Microsoft positions the family on price-performance, not frontier capability — that sentence is itself a positioning statement: it does not want MAI Code 2 judged against flagship models on absolute capability.
One sentence in the product post is easy to miss: both models are "explicitly not yet tuned" for harder problems, like fundamental science research and very-long-horizon agentic work. That is Microsoft's own scope limit, not a third party's criticism. Everyday enterprise coding — taking code to production, deep debugging, multi-file changes, repo-wide reasoning, UI work — is its home ground; beyond that, Microsoft itself does not vouch for it.
The MAI coding line did not appear from nowhere: per industry reporting, it started with MAI Code 1 Flash in June 2026, a lightweight coding assistant. MAI Code 2 is the line's first step from "lightweight assistance" to "complex, agentic software engineering" (GitHub's changelog wording). GitHub also announced that MAI-Code-1.1-Flash will retire across all Copilot experiences on Dec 18, 2026, and recommends switching to the latest MAI Code Flash model.
Reference: Microsoft AI — Product post "MAI Code 2 and Code 2 Flash: Everyday coding at scalable prices" · Microsoft AI — Ignite post "Seven new world-class models at scalable price-performance" · GitHub Changelog — MAI Code 2 and MAI Code 2 Flash in GitHub Copilot · Axios report (via superpowerdaily) (rewritten and organized by ModelBrief; facts per official sources)