Google has unveiled Gemma 4, its most intelligent open model family to date, purpose-built for advanced reasoning and agentic workflows, delivering an unprecedented level of intelligence-per-parameter. Since the launch of the first generation, developers have downloaded Gemma over 400 million times, fostering a vibrant Gemmaverse with over 100,000 variants.

Gemma 4 is released in four versatile sizes: Effective 2B (E2B), Effective 4B (E4B), 26B Mixture of Experts (MoE), and 31B Dense. These models can handle complex logic and agentic workflows, with the 31B model ranking #3 and the 26B model #6 on the industry-standard Arena AI text leaderboard, outperforming models 20x their size.

New capabilities include advanced reasoning, showing significant improvements in benchmarks requiring multi-step planning and deep logic, and native support for agentic workflows, enabling autonomous agents to interact with tools and APIs. It also features high-quality offline code generation, multimodal capabilities with native video and image processing, longer context windows (128K for edge, up to 256K for larger models), and native training on over 140 languages.

Gemma 4 is optimized to run and fine-tune efficiently on various hardware, from Android devices and laptop GPUs to developer workstations and accelerators. Specifically, the E2B and E4B models are engineered for on-device utility, running completely offline with near-zero latency across edge devices, bringing a new level of intelligence to mobile and IoT devices.

The models are released under a commercially permissive Apache 2.0 license, emphasizing developer flexibility and digital sovereignty. They are supported by a broad ecosystem of tools and platforms, including Hugging Face, Google AI Studio, Vertex AI, NVIDIA, AMD, and TPUs, and are built on a foundation of trust and safety.

👉 Read the full article on Google Blog

(Note: The original article was inaccessible. The following summary is based on the article's title and general industry trends.)

Microsoft is reportedly challenging its AI rivals, such as Google and OpenAI, by introducing three new foundational models. This move signifies a crucial step in the company's AI strategy, aiming to further expand the capabilities of generative AI.

These new foundational models are likely to focus on advanced reasoning capabilities, multimodal functionalities, and more efficient support for agentic workflows. Microsoft probably intends to offer a wide range of AI solutions for enterprises and developers through integration with its Microsoft Azure cloud services and existing products like Microsoft 365.

The release of these models underscores the rapid evolution of AI technology and the intensifying innovation race among major tech companies. Microsoft is presumed to be aiming to establish its leadership in the AI market and provide versatile AI solutions for diverse use cases through these new models.

👉 Read the full article on TechCrunch

(Note: The original article was inaccessible. The following summary is based on the article's title and general industry trends.)

News that a new Google-funded data center will be powered by a massive gas plant raises concerns about energy consumption and environmental impact associated with the rapid advancement of AI technology. Training and operating AI models require immense amounts of electricity, making data center energy sources a critical focus in sustainability discussions.

This report suggests that despite Google's stated commitment to renewable energy, new infrastructure may rely on fossil fuels, potentially highlighting a tension between the company's environmental goals and its real-world energy demands. The use of a large gas plant would lead to increased carbon emissions, casting doubt on efforts to combat climate change.

The article likely discusses the environmental footprint of AI growth, particularly how the choice of data center power sources impacts corporate sustainability strategies. This serves as an emblematic case of the challenge the entire technology industry faces in balancing innovation with environmental responsibility.

👉 Read the full article on Wired