Microsoft AI has introduced MAI-Thinking-1, a mixture-of-experts reasoning model with 35 billion active parameters, a 256,000-token context window, function calling, and Chat Completions API compatibility through Microsoft Foundry private preview.
Microsoft AI introduced MAI-Thinking-1 on June 2, 2026, describing it as the company’s first reasoning model and making it available through a Microsoft Foundry private preview.
In its announcement, Microsoft AI said MAI-Thinking-1 is designed for reasoning tasks that involve multi-step instructions, long-context work, and code generation. Microsoft News also described the model in its Build 2026 live blog as Microsoft AI’s first reasoning model, available to select early partners.
Microsoft AI’s model catalog page identifies MAI-Thinking-1 as a flagship reasoning model with 35 billion active parameters and about 1 trillion total parameters. The company describes the system as a mixture-of-experts model, a design that activates only part of the full model for a given request.
The announcement positions MAI-Thinking-1 as part of Microsoft AI’s broader model lineup rather than as a general release. According to Microsoft AI, access is currently through Microsoft Foundry private preview, with support for Chat Completions API compatibility.
Microsoft AI says MAI-Thinking-1 supports a 256,000-token context window, which is intended to help with long documents, extended conversations, and tasks that require the model to track information across large inputs. The company also lists function calling among the model’s supported capabilities, making it relevant for developers building applications that connect model responses to external tools or structured workflows.
The model’s Chat Completions API compatibility is also significant for developers already using that style of interface. Microsoft AI’s announcement says MAI-Thinking-1 is available through Microsoft Foundry private preview, rather than through a broad public launch.
Microsoft AI’s announcement cites results on SWE-Bench Pro and AIME, two benchmarks commonly used to evaluate coding and mathematical reasoning performance. The company’s model catalog page says MAI-Thinking-1 offers competitive reasoning and coding performance, though the source materials describe the results in the context of Microsoft’s own release materials.
Microsoft News’ Build 2026 live blog characterizes the model as built for multi-step instructions, long-context reasoning, and code generation. Together, the Microsoft AI announcement, model catalog page, and Build live blog present MAI-Thinking-1 as a developer-oriented reasoning model rather than a consumer product launch.
For now, Microsoft AI says MAI-Thinking-1 is available to select early partners through Microsoft Foundry private preview. The company has not described the sources provided here as a full public rollout.
The launch adds a Microsoft-branded reasoning model to the company’s AI portfolio, with the main disclosed technical details centered on its mixture-of-experts architecture, 35 billion active parameters, approximately 1 trillion total parameters, 256,000-token context window, function calling, and compatibility with the Chat Completions API.
Microsoft AI introduced MAI Thinking 1 on June 2, 2026, describing it as the company’s first reasoning model and making it available through a Microsoft Foundry private preview.
A reasoning model from Microsoft AI In its announcement, Microsoft AI said MAI Thinking 1 is designed for reasoning tasks that involve multi step instructions, long context work, and code generation.
Microsoft News also described the model in its Build 2026 live blog as Microsoft AI’s first reasoning model, available to select early partners.
Continue reading
