Live markets
BTC—— ETH—— SOL—— BNB—— XRP—— DOGE——
View all markets

Microsoft Introduces MAI-Code-2 and MAI-Cyber-2-Flash. Their Access Routes Are Different

Microsoft Introduces MAI-Code-2 and MAI-Cyber-2-Flash. Their Access Routes Are Different

Microsoft unveiled additions to its own AI model family at Ignite on November 17, including MAI-Code-2, MAI-Code-2-Flash and MAI-Cyber-2-Flash. The announcement emphasizes specialized models for software engineering and cybersecurity rather than relying on a single general-purpose model for every application.

The company is also expanding its Foundry model-routing and training tools, allowing developers to compare AI systems against the quality, cost, latency and data requirements of an actual workload.

Coding: flagship capability and a faster variant

MAI-Code-2 targets complex engineering jobs such as multi-file changes, repository-wide reasoning and difficult debugging. Code-2-Flash is tuned for common, high-volume coding work, including smaller fixes and explanations.

Microsoft says both coding models are generally available in GitHub Copilot, including the model picker and Copilot Fusion. Its detailed Foundry release notes separately describe a limited-access preview path for Code-2-Flash. Availability in one product should not be mistaken for unrestricted access in every Microsoft service.

Cybersecurity: a more specialized route

MAI-Cyber-2-Flash is designed for vulnerability discovery, validation and remediation. Microsoft says it was developed using its Codename MDASH security-testing environment and will be made available through that system. That future-oriented wording is different from a general release of the cybersecurity model to all Foundry customers.

The company also announced image-generation previews and improvements to model routing. Some capabilities are generally available, while others remain public preview or restricted offerings; teams should check their exact region and product eligibility before planning deployments.

Why smaller task-specific models matter

The strategy is to reserve heavy reasoning for complex requests while using cheaper, faster models for routine operations. This could lower average cost for enterprise AI systems, but model behavior still needs testing on representative codebases and security tasks.

The Microsoft release arrives alongside Google’s expansion of its Gemini model lineup. As AI tools move closer to production development, access controls, verification of code changes and clear responsibility for security findings remain as important as benchmark performance.

The important follow-up is adoption data: how well the coding models handle real pull requests, how reliably the security model distinguishes exploitable problems from false positives, and what developers actually pay per completed task. The launch sets out a broader product family, not proof that every workload improves automatically.

Source: Microsoft Foundry official announcement, November 17, 2026.