Microsoft drops six new MAI models at Ignite, cybersecurity joins lineup
Microsoft rolled out six new MAI models at Ignite, covering coding, cybersecurity, image generation and voice — with security as the brand-new category. It follows the seven-model batch unveiled at Build 2026, headlined by MAI-Thinking-1, which Microsoft says is its first advanced reasoning model built from scratch.
- Original source: cryptobriefing.com.
- The source published it on Nov 17, 2026 at 6:23 PM.
- No second outlet has reported it yet. Treat it as a single-source claim.
The six new models sit under the Microsoft AI banner, or MAI, and split across four kinds of work: writing code, defending systems, generating images and producing voice. Cybersecurity is the fresh face here — earlier MAI releases covered reasoning, coding, images, transcription and voice, but not security.
It's an add-on to a bigger drop from earlier in the year. At Build 2026, held June 2 in San Francisco, Microsoft introduced seven MAI models, among them MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Transcribe-1.5 and MAI-Voice-2. Some arrived with lighter Flash variants built to trade capability for speed and lower cost.
MAI-Thinking-1 headlined that Build batch. Microsoft billed it as its first advanced reasoning model, built from scratch on clean, commercially licensed data rather than leaning on outside models like OpenAI's. It runs 35 billion active parameters with a 256K context length, scored 97% on the AIME 2025 competition math benchmark, and held up favorably against Anthropic's Claude Sonnet 4.6 in blind evaluations, according to Microsoft.
Mustafa Suleyman, Microsoft's AI chief, positioned the family as a real contender next to Anthropic's Claude line and Google's image models. The pitch leans on efficiency, enterprise-grade data practices and tight hooks into tools Microsoft already owns — GitHub Copilot, Visual Studio Code and Microsoft Foundry.
Ignite 2026 runs November 17 to 20, 2026, with dedicated MAI sessions on the schedule. Microsoft keeps stressing that the models are trained on clean data and built without third-party distillation — the trick of training a smaller model on a larger one's outputs — which lets it call them genuinely its own.
TopicsMicrosoft · MAI · Mustafa Suleyman · MAI-Thinking-1 · Anthropic · Claude Sonnet 4.6 · GitHub Copilot · Google
Source: cryptobriefing.com
Written by AI from verified reporting.
More stories
- Tech & AI cryptobriefing.com 0 viewsMicrosoft AI Unveils MAI Image 2.7, Challenging Image Editing Leaders
- Tech & AI breakingthenews.net 1 viewMicrosoft and Salesforce unveil Copilotforce at Ignite conference
- Tech & AI techcrunch.com Caught by @hafiz338 0 viewsGoogle Opens Gemini Tasks to Hundreds of Millions, Toggle Turns Chatbot Into Errand Runner
- Tech & AI cryptobriefing.com 1 viewAnthropic IPO order book shuts at 4 p.m. ET, reportedly six times oversubscribed

