Google DeepMind has announced its next frontier model, Gemini 4 Carbon, which stands out for its state-of-the-art cybersecurity and long-horizon capabilities. The model demonstrates significant advancements in coding, agentic tasks, and cybersecurity defense, achieving top rankings in key benchmarks such as SEC-bench Pro, ExploitGym, and CWE-bench.

Gemini 4 Carbon is engineered for sustained agentic workflows, capable of tackling complex, multi-step engineering problems like root-causing and patching subtle bugs across large codebases over several days. Emphasis has also been placed on enhancing model safety and alignment, with improved resistance to prompt injection.

The rollout of Gemini 4 Carbon will be deliberate, initially made available to cyber defenders and trusted partners, before a broader release to developers, businesses, and consumers. Pricing during the introductory period will match Gemini 4 Argon, at $2 per million input tokens and $10 per million output tokens.

👉 Read the full article on Google Blog

Apple has unveiled a groundbreaking new MacBook Pro featuring the innovative PROLED XDR display. This display incorporates revolutionary tandem OLED technology and, for the first time on a Mac, supports intuitive touch operation.

The new MacBook Pro is powered by the new M5 Pro and M5 Max chips, delivering exceptional performance and robust on-device AI capabilities. Furthermore, the Dynamic Island makes its debut on Mac, serving as a new home for Spotlight, Live Activities, and Siri AI.

The design is thinner and lighter, offering up to 24 hours of battery life. This device is tailored for creative professionals and developers, leveraging the power of Apple Intelligence and macOS 27 Golden Gate to introduce new dimensions in content interaction, from manipulating 3D models to playing virtual instruments directly on screen.

👉 Read the full article on Apple (日本)

Google DeepMind has introduced its latest AI world model, Genie 4, which is capable of generating photorealistic videos. Unlike its predecessor, Genie 3, this new model can also automatically generate audio to complement the visual scenes it creates.

A significant differentiator for Genie 4 is its enhanced ability to simulate humans, featuring more fluid bodily motions and facial expressions. Crucially, it can also generate human voices, allowing AI-generated characters to converse with each other within the video.

The article highlights world models as a major investment area for achieving Artificial General Intelligence (AGI), with experts like Yann LeCun suggesting they might be a more promising path than LLMs. Genie 4 is being rolled out with strict safety guardrails and SynthID watermarking, initially to early testers, including video game developers for applications like "vibe gaming."

👉 Read the full article on Gizmodo