MAI-Code-2 is Microsoft's first-party coding model, announced at Ignite 2026 and generally available in GitHub from day one of the conference. Microsoft positions it for demanding agentic coding work, as distinct from MAI-Code-2-Flash, which targets high-volume everyday coding and is also in private preview in Microsoft Foundry.
It arrived alongside five other models: MAI-Image-2.7 and MAI-Image-2.7-Flash for image generation, MAI-Cyber-2-Flash for vulnerability discovery and remediation, and Microsoft-Decision-1.1 for routing and classification. Of the six, only the two coding models and Microsoft-Decision-1.1 are generally available; the rest are preview.
The significant gap is pricing. Microsoft's Ignite Book of News gives no per-million-token cost for any model in the announcement, despite widespread coverage framing the launch as a cheaper alternative to Claude and GPT-6. Access runs through GitHub, so in practice cost is governed by your Copilot arrangement until Microsoft publishes a rate.
Weights are closed. No independent benchmark results were published with the launch.