Microsoft unveiled seven new AI models at its Build conference, with the flagship being the 35-billion-parameter MAI-Thinking-1. This model is touted as Microsoft's "first reasoning model" and reportedly outperforms Anthropic's Claude Sonnet 4.6 in overall quality. It achieved a 97% on the AIME benchmark and 53% on SWE Bench Pro for complex coding tasks, surpassing Claude Opus 4.6 but trailing OpenAI's GPT-5.4.

A key selling point for MAI-Thinking-1 is its training "entirely from the bottom" with "absolutely zero distillation," meaning it wasn't trained using other companies' models. Microsoft is leveraging legal uncertainties and copyright infringement concerns surrounding AI training data to market this model to businesses, emphasizing its "enterprise-grade, clean, and commercially licensed data lineage" for trustworthy production use.

Other models introduced include image-generating models MAI-Image-2.5 and MAI-Image-2.5 Flash, transcription model MAI-Transcribe-1.5, speech-generation models MAI-Voice-2 and Mai-Voice-2 Flash, and code-generating model MAI-Code-1-Flash. This announcement marks Microsoft's largest in-house AI product release since August, indicating a growing independence from its previous heavy reliance on OpenAI.

👉 Read the full article on Gizmodo

Google has partnered with the Utah State Board of Education (USBE) to provide Gemini for Education to all K-12 schools across Utah, starting in the 2026-2027 school year. This initiative offers secure AI tools, specialized training, and free access to Google Career Certificates for over 708,000 students and educators.

For educators, Gemini aims to streamline administrative tasks and enhance lesson planning by generating customized materials like assignments and grading rubrics, freeing up time for direct student mentorship. Students will benefit from personalized assistance to explore complex concepts, generate creative ideas, and receive tailored explanations through tools like Guided Learning.

A core principle of these educational tools is data privacy; content and conversations within Gemini for Education are private, protected with enterprise-grade security, and are not used to train Google's AI models. High school students (grades 9-12) will also receive free access to Google Career Certificates and AI training until December 2027, equipping them with skills for high-demand jobs in fields such as cybersecurity and data analytics.

👉 Read the full article on Google Blog

Solid fuel rocket engines, while reliable and powerful, have long suffered from a critical limitation: once ignited, their combustion cannot be stopped until all fuel is consumed. This inflexibility poses significant challenges for modern space missions requiring precise trajectory corrections or orbital adjustments.

Researchers from Aerospace Corporation, the University of Southern California, and the U.S. Naval Postgraduate School have developed a novel system that could overcome this problem. Their innovation combines a special electricity-conducting ionic liquid polymer fuel with a technique called nanosecond pulsed plasma discharge.

By applying brief, high-voltage electrical pulses to generate plasma near the combustion zone, electrons and free radicals are produced, which can modify the chemical reactions sustaining the flame. Laboratory results demonstrate that this method can slow down, stop, or reactivate combustion, effectively allowing for on-demand control of solid fuel engines. This breakthrough, though still a proof of concept, holds particular promise for small satellites and CubeSats, offering the economic and mechanical advantages of solid propulsion with unprecedented flexibility.

👉 Read the full article on Gizmodo en Español