Microsoft AI announced a new set of MAI models, led by the MAI-Thinking-1 reasoning model and the MAI-Code-1-Flash coding model. Microsoft describes MAI-Thinking-1 as a 35B-active sparse mixture-of-experts model trained on commercially licensed data, while DataNorth reports a 256K-token context window and listed pre...
Microsoft AI has announced seven new MAI models, including MAI-Thinking-1 and MAI-Code-1-Flash, as part of a broader expansion of its in-house model lineup.
In a post titled “Building a hill-climbing machine: Launching seven new MAI models,” Microsoft AI said the release includes seven MAI models. The company described MAI-Thinking-1 as its flagship reasoning model and MAI-Code-1-Flash as an inference-efficient model for agentic coding.
Microsoft AI said MAI-Code-1-Flash is integrated with GitHub Copilot, Visual Studio Code, and the Microsoft software stack. That positioning places the coding model in developer workflows where Microsoft already has major distribution through GitHub and VS Code.
In a separate post, “Introducing MAI-Thinking-1,” Microsoft AI described MAI-Thinking-1 as a reasoning model built as a 35B-active sparse mixture-of-experts model. Microsoft AI said the model was trained on clean, commercially licensed data and without third-party distillation.
DataNorth’s report on the launch says MAI-Thinking-1 has a 256K-token context window. The same report says Microsoft cited results on AIME and SWE-Bench Pro, though the source summary does not provide full benchmark tables or independent verification details.
The emphasis on commercially licensed training data is notable because model provenance has become a central issue for AI developers, customers, and rights holders. Microsoft AI’s own description frames MAI-Thinking-1 as part of its effort to build internal model capabilities rather than relying only on external providers.
Microsoft AI described MAI-Code-1-Flash as an inference-efficient agentic coding model. According to Microsoft AI, the model is tied into GitHub Copilot, VS Code, and Microsoft’s broader stack, suggesting a focus on code generation, code editing, and software development assistance inside existing Microsoft tools.
DataNorth reports listed pricing for MAI-Code-1-Flash at $0.75 per million input tokens, $0.075 per million cached input tokens, and $4.50 per million output tokens. DataNorth also noted that pricing was still being finalized, so those figures should be treated as launch-period information rather than final long-term pricing.
The available sources establish that Microsoft AI announced seven MAI models, that MAI-Thinking-1 is presented by Microsoft AI as a reasoning model, and that MAI-Code-1-Flash is positioned for coding use cases across Microsoft developer products.
The sources also state that MAI-Thinking-1 is a 35B-active sparse mixture-of-experts model trained on commercially licensed data, and that DataNorth reported a 256K-token context window plus preview-style pricing details for MAI-Code-1-Flash.
Microsoft AI’s announcements show a continued push to build proprietary model infrastructure across reasoning and coding. However, based on the provided sources, claims about real-world performance should be limited to Microsoft’s and DataNorth’s reported descriptions rather than treated as independently validated conclusions.
Microsoft AI has announced seven new MAI models, including MAI Thinking 1 and MAI Code 1 Flash, as part of a broader expansion of its in house model lineup.
Microsoft expands the MAI model family In a post titled “Building a hill climbing machine: Launching seven new MAI models,” Microsoft AI said the release includes seven MAI models.
The company described MAI Thinking 1 as its flagship reasoning model and MAI Code 1 Flash as an inference efficient model for agentic coding.
Continue reading
