Particle.news

Microsoft Unveils MAI Code 2 Models for Copilot and a Cybersecurity MAI

The rollout signals Microsoft’s push to match rival model performance while cutting cloud-run costs and tightening safety controls for enterprise users.

Overview

  • Microsoft deployed MAI Code 2 and MAI Code 2 Flash into GitHub Copilot on Tuesday, with Code 2 Flash also placed in a limited Microsoft Foundry preview.
  • The company released a suite of new MAI models covering coding, cybersecurity, image generation, transcription and voice at its Ignite conference on November 17, 2026.
  • Microsoft and a partner, AT&T, reported tests that show Code 2 Flash outperformed certain GPT-6.1 variants and ran far cheaper in vendor benchmarks, but independent verification of those claims is still pending.
  • The firm introduced a cyber-specific MAI model for select customers and said model alignment and safeguards are a top priority after recent AI-related security incidents that exposed platform vulnerabilities.
  • Microsoft says the MAI family is built from scratch on commercially licensed 'clean' data and tuned for task-specific efficiency, a strategy meant to lower operating costs for developers and large enterprises while keeping control over data and pricing.