Microsoft released MAI-Image-2.7-Preview on October 27, a controllable image editing model that takes an original image and text instructions. It ranks second on Artificial Analysis's image editing leaderboard with 1,193 Elo, behind OpenAI's GPT Image 2.5 Sunburst (max). The preview currently offers image editing only; no API, pricing, or commercial release date has been announced.
On October 15, 2026, Microsoft AI CEO Mustafa Suleyman said at an employee meeting that Microsoft's in-house pretrained foundation model MAI-Base-2 outperforms OpenAI's model powering GPT-5.6 Sol, using only a quarter of the training compute and one-twelfth of DeepSeek's compute. Microsoft released the model earlier that day and had privately compared it with several OpenAI models on a confidential basis.
Microsoft announced that its next-generation in-house AI inference chip, Maia 300, will be deployed at gigawatt scale in data centers in 2027, delivering about 40% higher token output per megawatt than mainstream GPU systems, and will prioritize internal AI services rather than renting out compute capacity.
On October 8, Helm.ai announced that it had signed $70 million in commercial contracts over the past 12 months, with customers spanning global automotive OEMs, Tier 1 suppliers, and industrial automation companies. The business has expanded from automotive into industrial automation and robotics, and the company is moving toward operating breakeven.
Microsoft announced at its San Francisco event that GitHub Copilot will support on-device AI model inference by the end of this month, allowing developers to switch between cloud and device-side models automatically or manually, and to set preferences in CLI, Copilot app, and VS Code. It also introduced MAI Code 1.1 Flash, a mixture-of-experts model with 137B total parameters and 6.8B active parameters, achieving 40-63 tokens per second decode throughput on Surface Laptop Ultra.
Microsoft released its first real-time streaming speech-to-text model MAI-Transcribe-2-Streaming, supporting 60 languages and automatic language detection, priced at $0.54 per hour during the promotional period, accessible via Microsoft Foundry, MAI Playground, and OpenRouter. In Artificial Analysis tests, it achieved a final transcription latency of 0.13 seconds and a word error rate of 2.50%, ranking first among 28 streaming speech-to-text models.
Microsoft AI's MAI-Voice-2.1 and MAI-Voice-2.1-Flash speech models are now available on OpenRouter, supporting the same voice across 23 languages with native accents, priced at $22 and $15 per million characters respectively, for real-time voice and audio content use cases.
Anthropic has signed a seven-year, $11.6 billion compute agreement with Akamai, under which Akamai will provide CPU compute and issue warrants, with a potential further expansion of up to $9 billion for a total potential scale of about $20 billion.
Microsoft CEO Satya Nadella outlined Microsoft's frontier AI philosophy on X, emphasizing that AI should develop under core principles, accelerate progress and benefit all parties. He voiced support for AI alignment research and independent audits, and announced that Microsoft will release its conduct guidelines for its in-house MAI models for public feedback on September 15.