Anthropic has introduced its latest AI models, Claude Fable 5.1 and Claude Mythos 5.1, positioning them as the world's most advanced for coding and knowledge work. Their research capabilities offer an early glimpse into how AI models will contribute to scientific progress.

Fable 5.1 is generally available, while Mythos 5.1 is offered through trusted access programs with enhanced safeguards, specifically designed for cybersecurity and life sciences. These models achieve significant cost reductions, up to 25% for typical workloads and up to 45% for highly agentic tasks, compared to Fable 5.

New Enterprise Frontier Safeguards (EFS) provide customers with complete privacy, equivalent to a zero-data retention policy. Fable 5.1 also features improved safeguards, reducing false positives by 60% and enabling software vulnerability discovery. Mythos 5.1 demonstrates remarkable advancements in molecular design, creating high-affinity binders, and in computational biology, optimizing GPU kernels to accelerate research and reduce costs.

In terms of safety and alignment, the models underwent extensive testing for chemical, biological, cyber, and agentic risks. Mythos 5.1 shows improved alignment and robustness against malicious requests and prompt injections compared to its predecessor. In compliance with the EU AI Act, a new invisible watermark feature has been implemented to ensure transparency of AI-generated content.

👉 Read the full article on Anthropic

Google has launched agentic video understanding for its Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite models. This new capability allows the models to dynamically scan video segments, significantly improving accuracy while reducing token consumption by up to 88% and costs by up to 66%.

Unlike traditional "static" processing, agentic video understanding combines Gemini's core reasoning with native video tools to dynamically search, scan, and inspect target video segments across visual frames, audio, and transcripts. This agentic loop allows Gemini to autonomously load relevant video parts, drastically cutting development overhead that previously required manual intervention.

The feature transforms how developers can process long-form video content, enabling capabilities such as sub-second moment retrieval, "needle-in-a-haystack" search across multi-hour videos, precise anomaly detection, and accurate counting of actions and objects. These efficiency gains are particularly impactful for long-form videos, where static processing often forced a trade-off between high token costs and missing critical details.

Agentic video understanding is available today via the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform, utilizing standard Gemini API token pricing without additional feature fees. Google plans to roll out this feature to billions of users across Google products, including the Gemini app and YouTube's 'Ask YouTube' feature, to deliver higher-quality, visually grounded answers.

👉 Read the full article on Google Blog

(Based on inferred information from the title, as the article content was inaccessible.)

Google has further solidified its commitment to clean energy by signing an agreement to purchase 400 megawatts (MW) of enhanced geothermal energy from Fervo Energy. This move is part of Google's ongoing efforts to power its data centers and operations with renewable energy sources.

Enhanced geothermal systems offer a sustainable and reliable source of baseload power, available in a wider range of geographical locations than traditional geothermal. Google's substantial purchase from Fervo signals confidence in the commercial viability of this technology and its role in achieving corporate decarbonization goals.

This agreement represents a significant step towards Google's ambitious goal of operating its businesses on 24/7 carbon-free energy by 2030. It also highlights the role major tech companies play in accelerating the development and adoption of geothermal energy technologies, particularly enhanced geothermal.

👉 Read the full article on TechCrunch