Anthropic has introduced the Model Hardware Standard (MHS), a new specification designed to let AI agents safely operate laboratory and manufacturing equipment. The system can coordinate devices such as microscopes, robotic arms and liquid handlers through a common interface.anthropic.com ↗
IBM released Canadian findings from its latest sports technology survey. According to the company, 68% of surveyed fans see value in AI-powered sports features, while 54% want faster event recaps and 51% are looking for more personalized experiences as digital platforms become more central to sports viewing.
Google released Gemini 3.5 Transcribe for developers. According to the company, the speech-to-text model improves accuracy in noisy audio, technical vocabulary and natural conversation, supports sub-second latency for live streaming, and can process recorded audio with speaker attribution and word-level timestamps.blog.google ↗
Cohere introduced Parse, a model for converting complex enterprise documents into structured data. According to the company, it goes beyond text extraction by understanding tables, forms, diagrams and visual elements in context, preserving document layout and producing output suitable for search, RAG and agentic workflows.cohere.com ↗
Google introduced Expert Intelligence for Gemini Notebook. According to the company, users can add eligible ebooks they purchased through Google Play Books directly to a notebook, letting Gemini answer questions grounded in those sources and generate cited explanations, infographics, audio overviews and quizzes from the material.blog.google ↗
Anthropic expanded Claude access for scientists with a new 10,000-seat program. According to the company, standard team seats will be free for one year, while premium seats with five times the usage limits will cost $15 per month. Anthropic says it plans to extend the program beyond the initial allocation over the coming months.anthropic.com ↗
Google introduced the Planetary Prediction Engine, which automates geospatial modeling from a natural-language question through data discovery, cleanup, feature engineering, model training and reporting. According to the company, it beat manually built expert pipelines on several tests while reducing work that can take weeks to minutes.research.google ↗