MAI-Thinking-1: Microsoft's New Reasoning AI Model Unveiled

Microsoft has unveiled new in-house AI models, including its first reasoning model called MAI-Thinking-1. This model aims for high efficiency at a low cost. MAI-Thinking-1 is currently in private preview through the Microsoft Foundry platform. It reportedly matches Anthropic’s Claude Opus 4.6 model in coding abilities on the SWE Bench Pro benchmark. What's interesting is Microsoft also introduced MAI-Image-2.5, a text-to-image and image-to-model. It's available in PowerPoint and Microsoft Foundry, and will soon be in OneDrive. This model is said to surpass Google’s Nano Banana Pro model on the ELO benchmark. Additional updates include MAI-Transcribe-1.5, offering high accuracy across 43 languages, and MAI-Voice-2, with over 15 new languages and voice options. MAI-Code-1, an inference coding model, is now integrated into GitHub Copilot and VS Code. The bottom line is these new Microsoft AI models will soon be accessible on more platforms, including Fireworks AI, Baseten, and OpenRouter.