Microsoft releases MAI-Thinking-1, its first reasoning model with 35B parameters
Microsoft released seven AI models at Build 2026, headlined by MAI-Thinking-1, its first reasoning model with 35 billion parameters. The company claims the model matches Anthropic's Claude Opus 4.6 on SWE Bench Pro coding benchmarks and beats Sonnet 4.61 in blind tests.
Microsoft releases MAI-Thinking-1, its first reasoning model with 35B parameters
Microsoft launched seven AI models at its Build 2026 developer conference, including MAI-Thinking-1, the company's first reasoning model.
MAI-Thinking-1 specifications
The 35-billion-parameter model was trained on "enterprise-grade, clean and commercially licensed data," according to Microsoft. The company claims it beat Anthropic's Claude Sonnet 4.61 when evaluated by independent reviewers in blind tests, and matches Claude Opus 4.6's SWE Bench Pro benchmark score for coding. Specific benchmark numbers were not disclosed.
MAI-Thinking-1 is designed for multi-step agentic tasks and is available in private preview through Microsoft Foundry. Context window size and pricing have not been announced.
Six additional models
MAI-Code-1: A coding model Microsoft describes as "ultra-efficient" and "tuned for GitHub." Available today in Copilot and VS Code. No parameter count or benchmarks disclosed.
MAI-Image-2.5 and MAI-Image-2.5 Flash: Microsoft's first text-to-image and image-to-image models. According to Microsoft, MAI-Image-2.5 outperformed Nano Banana Pro on ELO ratings and ranked third on the LM Arena Leaderboard at launch, behind Nano Banana. Available now in PowerPoint, Foundry, and rolling out in OneDrive.
MAI-Transcribe-1.5: A speech transcription model supporting 43 languages, with streaming capabilities coming soon. This represents an update to the MAI-Transcribe line released just two months prior.
MAI-Voice-2 and MAI-Voice-2 Flash: Text-to-speech models supporting 15 more languages than the previous MAI-Voice-1, which was released in preview two months ago.
Microsoft AI CEO Mustafa Suleyman stated that all models include watermarking "from scratch" and offer cost efficiency improvements up to 10x compared to competitor models, though specific pricing was not provided.
Availability and healthcare partnership
All MAI models will be available through Fireworks AI (now generally available on Foundry), Baseten, and OpenRouter.
Microsoft also announced a partnership with Mayo Clinic to develop a frontier model for healthcare, joining existing medical AI efforts from OpenAI and Google. Microsoft already offers Copilot Health.
What this means
Microsoft's entry into reasoning models puts it in direct competition with OpenAI's o1 series, Anthropic's Claude family, and DeepSeek's R1. The emphasis on commercially licensed training data addresses enterprise customers' copyright concerns, though the lack of disclosed pricing and limited benchmark data makes performance comparisons difficult. The rapid release cycle—with voice and transcription model updates arriving just two months after initial previews—signals Microsoft's aggressive push to compete across the full AI model stack.
Related Articles
Chariot launches Chariot-1-ASR Realtime, a streaming speech recognition model for 10 Indian languages
Chariot, a frontier AI research lab, launched Chariot-1-ASR Realtime on October 22. The streaming speech recognition model covers 10 Indian languages plus Indian English. The company claims it made fewer errors than every streaming and batch system it tested, based on its own evaluations.
Microsoft's ThinkingBox-Training lifts Qwen3.8-27B fully reliable task completions from 38 to 107 of 507
Microsoft and Hugging Face released ThinkingBox-Training, which turns executable database-state checks into RL rewards. After 50 GRPO steps, Qwen3.8-27B's pass@1 rose from 51.70% to 60.78%, and tasks passed on all 20 attempts rose from 38 to 107, according to the authors.
MOSS-Image, a 5B open-weights image model, passes one million downloads in four days
MOSS-Image, an open-weights text-to-image model from MOSS (模思智能) and the OpenMOSS team, is built on a 5B-parameter DiT backbone and has passed one million downloads in four days, according to Chinese outlet QbitAI. It supports structured JSON layout input and text prompts up to 4096 tokens.
Kimi K3.1 Pro launches with 1M-token context, 65.7% on Terminal-Bench 4.0; open weights due by Nov 4
Kimi K3.1 Pro, the first model in the K3.1 series, is a 3T-class open-weight model with a 1-million-token context window. It scores 65.7% on Terminal-Bench 4.0 and is available today on Kimi.com, Kimi Work, Kimi Code and the Kimi API. Full weights will be released by Nov 4.
Comments
Loading...