Microsoft on Tuesday released a family of seven in-house AI models at its Ignite conference in San Francisco, led by a coding model it says matches most of OpenAI's and Anthropic's performance for a small share of the price. The company framed the launch around agents. "As token spend skyrockets, we know that you also want models at a price point that lets you run agents 24-7 doing real engineering work," Microsoft AI wrote in its announcement.
The lineup comes from Microsoft AI, the group known as MAI. It spans coding, cybersecurity, image editing, real-time voice, medical transcription and model routing.
Coding models built for agent workflows
MAI Code 2 is aimed at debugging, multi-file implementation, refactoring and repository-wide reasoning. Averaged over four benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1), Microsoft says it delivers 98% of the performance of OpenAI's GPT-6.1 Sol at 40% of the price, and 96% of Anthropic's Claude Opus 5.5 at 5% of the price.
The company also said developers preferred MAI Code 2 over GPT-6 Sol 64% of the time in a blind, side-by-side evaluation of frontend work.
MAI Code 2 Flash is a smaller model for code explanations, targeted edits, routine fixes and rapid iteration. Microsoft says it performs comparably to GPT-6 Luna at 65% of the price and Claude Haiku 5.5 at 20% of the price across three of those benchmarks.
Microsoft was direct about the limits. The models are "explicitly not yet tuned for harder problems, like fundamental science research and very long-horizon agentic work," the MAI Code 2 announcement says. All of the benchmark comparisons come from Microsoft and had not been independently replicated at publication.
Where developers can use them
GitHub said both models are now in Copilot and are "designed for agentic coding workflows." MAI Code 2 is available on Copilot Pro, Pro+, Max, Business and Enterprise plans, while Code 2 Flash also reaches Free and Student users, according to the GitHub changelog. The models appear in the model picker in VS Code, Visual Studio, the Copilot CLI and the Copilot cloud agent, among other surfaces, and are billed under usage-based pricing.
New models are enabled by default for Business and Enterprise customers unless an administrator turns that off, GitHub said. Code 2 Flash is also in a limited-access preview in Microsoft Foundry. Microsoft said it plans to bring both models to OpenRouter, Vercel, OpenCode and Kilo, among others.
Axios reported that the models could shift some GitHub Copilot workloads away from outside providers such as OpenAI. Mustafa Suleyman, who leads Microsoft AI, declined to give Axios a target for how much Copilot traffic would move to Microsoft's models but predicted significant adoption.
Security agents and routing
Microsoft said security and containment are its top priorities for agent development, and that defenders need "monitoring and control agents defending and patching live production systems 24-7." It said MAI Cyber 2 Flash beats GPT-6 Astra on the CyberGym-E2E benchmark at less than 2% of the cost per task. Earlier MAI Cyber models account for almost 40% of scans in MDASH, its security scanning service, and the new model is priced 86% lower on output tokens than MAI Cyber 1.1 Flash, according to the company.
Decision 1.1, a routing model meant to send work to larger models only when needed, is used by 4,000 customers, Microsoft said. It described the model as 50 times faster and 79 times cheaper than GPT-6.1 Sol.
The company also introduced MAI Licensed Weights, which let enterprises customize a MAI model with their own data and run it in their own Azure environment. Microsoft said AT&T built a custom MAI model for 5G network analytics that outperformed GPT-6.1 Sol at one-fiftieth of the cost.
Source: Microsoft AI, “Seven new world-class models at scalable price-performance”
Additional sources: Microsoft AI, “MAI Code 2 and Code 2 Flash: Everyday coding at scalable prices”; GitHub Changelog, “MAI Code 2 and MAI Code 2 Flash in GitHub Copilot”; Axios, “Microsoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI models”




