Microsoft's new models belatedly target the same coding and security domains as models from OpenAI and Anthropic. They address enterprise concerns about cost and routing the right model to the right workload.
The cloud provider launched Decision 1.1, built on Google DeepMind's open-weight model, Gemma, at its Microsoft Ignite 2026 conference on Tuesday. It supports fast and what Microsoft calls cost-effective routing, as well as classification, prioritization and verification and helps developers provide guidelines for the model to make the right decision for each AI workload.
Microsoft also expanded its MAI model family. MAI-Code-2 is for demanding agentic coding, while MAI-Code-2-Flash supports high-volume everyday coding. Both are available on GitHub. Code-2-Flash is in private preview in Foundry, Microsoft's AI platform for building and scaling AI applications and agents. MAI-Image-2.7 and MAI-Image-2.7-Flash are in preview and target sharper editing and consistent image generation. MAI-Cyber-2-Flash, a cybersecurity model available in preview, helps security teams find, confirm and fix vulnerabilities. Microsoft said "trusted partners" will get first access to MAI model weights.
"We're committed to doing what it takes to ensure that everyone can own their own intelligence," chairman and CEO Satya Nadella said during Tuesday's keynote.
With its new models, Microsoft is looking to show enterprises that it can stand on its own after ending its exclusive agreement with OpenAI in April, focusing on enterprise challenges such as model pricing and routing.
Price war and routing
Microsoft's new models are significantly cheaper than competitors'. MAI-Code-2 standard pricing is $1 per million input tokens and $5 per million output tokens. Comparatively, GPT-6.1 from OpenAI costs $2 per million input tokens and $10 per million output tokens. Anthropic’s Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens.
"With this release, the MAI family has matured into models that cost less and still hold up against the other big model makers," said Michael Heath, principal architect, cloud and AI, at Myriad360, a global systems integrator and Microsoft partner. "Microsoft is competing on price performance more than on raw intelligence."
Microsoft is also focusing on helping enterprises find the right model for each use case. The model router in Foundry automatically does that.
"For our clients, the useful question has moved from ‘which model?’ to ‘which platform lets me swap models without rebuilding my agents?’" Heath said, adding that the ability to route to models automatically lets enterprises focus on "outcomes instead of picking a model or watching the meter."
Much to prove
However, Microsoft still needs to show that the operational systems built around its models are beneficial for enterprises.
"The question is whether its models give customers an advantage inside the products they already use," said Mitch Ashley, an analyst at Futurum Group. "Microsoft doesn't have to win the standalone model race if it can deliver better economics and useful capabilities through GitHub Copilot, Foundry and its security portfolio."
Ashley added that Microsoft's pricing also needs to make sense for each application. For instance, while Cyber 2 Flash is a lower-cost, practical, specialized model for agent-driven security operations, its performance needs to be validated independently by enterprises.
"In security, cheaper is not a better answer if it misses an attack or triggers the wrong remediation," he said. The same applies to Code 2 and Code 2 Flash, he added.
"Enterprises will judge it on the quality and total cost of completed work, and token price is only one input to that number," Ashley continued.
Esther Shittu is a TechTarget reporter and podcast host and producer covering AI systems.