Don’t rent a generic model.
Tune your own.
Tune your own.
Cheaper than GPT-6.1 Sol for 5G network analysis, while beating it on held-out tests.
Of MISRA fixes that needed human rework from frontier models, now merge-ready on the first try.
Accuracy in classifying aircraft reports, beating GPT-6.1 Sol at a fraction of the cost.
Learn more about Licensed Weights
MAI Licensed Weights will initially be available through our Frontier Company engagements. To learn more, contact your Microsoft account team.
Featured partners
“Our engineers run one of the most advanced 5G networks in the country, and we want to give them the best tools available. For standard 5G network analysis, our custom version of Code 2 Flash outperforms GPT-6.1 Sol, while being 50x cheaper. For more complex workflows, we found that Code 2 delivers frontier performance, at a significantly lower price than alternatives.
“In pursuit of a zero-accident future, the automated-driving software we develop follows MISRA, an automotive industry coding standard for reliable embedded software. Out of the box, MAI Code 2 Flash is a powerful, efficient agentic coding model. After fine-tuning on our MISRA fix history, it delivered ready-to-merge fixes for 44% of cases where fixes from the frontier models we run in production needed human rework, at less than one-tenth of the cost. That validates the hill-climbing loop we set out to build, and this is only the first step up the hill.
“Capturing customer intent across languages, dialects, and nuanced conversations is a complex challenge. Fine-tuning MAI Code 2 Flash delivered a significant improvement in accuracy in our evaluations, while retaining the efficiency that makes it ideal for high-volume workloads.