Google Translate now integrates the Gemini 2.5 Flash Native Audio model, supporting real-time voice translation for over 70 languages, compatible with all headphones, and retaining the speaker's tone and pace, significantly enhancing cross-language communication experience. Furthermore, Google has expanded its language practice mode and pronunciation feedback, promoting the widespread adoption of AI in global communication.

Zoom announced its 'Federated' AI system scored 48.1% on the Humanity's Last Exam reasoning benchmark, surpassing Gemini 3 Pro (45.8%) and second only to GPT-5.2 (50%). This system dynamically orchestrates models from OpenAI, Anthropic, Google, and others through a 'Z-scorer', establishing a new paradigm for enterprise-level multi-model AI collaboration.

OpenAI officially launched the GPT-5.2 series, targeting professional expertise scenarios. On the GDPval benchmark, it surpassed human experts on 71% of tasks, demonstrating significant improvements in hallucination rate, visual understanding, coding capability, and deep reasoning, positioning it as a 'Code Red' strategic product to counter Google Gemini 3.

Runway introduced the GWM-1 General World Model, based on the Gen-4.5 architecture. It supports multi-minute interactive video generation at 24FPS 720p resolution, allowing real-time manipulation of virtual environments through multimodal inputs like actions, camera controls, and audio. This aids in robotics policy training and digital twins, marking a shift from generative AI to simulation AI.

Oracle's market value plummeted by $80 billion following a sharp quarterly capital expenditure surge to $12 billion, missed revenue expectations, and negative free cash flow. This exposes the physical bottlenecks and ROI pressures in AI infrastructure buildout, prompting a market shift from 'GPU hoarding' to focusing on actual data center commissioning and cash flow discipline.

Founded by the former head of AI at Databricks, Unconventional AI raised $475 million just two months after its inception, achieving a $4.5 billion valuation. It aims to develop a new, brain-inspired AI computing platform to address AI energy consumption bottlenecks, reflecting heightened industry concern over the 'compute-power' crisis.

Google DeepMind will establish its first automated AI science laboratory in the UK. It will use AI to accelerate the development of new materials (e.g., superconductors, semiconductors, solar) and will provide model access to UK scientists and AI safety research institutes, promoting the application of AI in public services like education and energy.

The Tinker API is now fully available. It introduces new capabilities like the Kimi K2 Thinking reasoning model and Qwen3-VL visual input, is compatible with the OpenAI API, supports mixed image-text reasoning and efficient fine-tuning, empowering enterprises and developers to build multimodal AI applications.

AI development tools like Claude Code enhance the efficiency and accuracy of tasks such as code generation and project migration (e.g., CMS to Markdown) through boundary-aware queues, planning modes, and memory systems. AI-assisted development is gradually evolving from a 'toy' into a practical productivity tool.

Traditional IAM struggles to track the dynamic permission chains of AI agents. Capability-based delegation mechanisms, using cryptographic tokens to enforce minimum privileges and traceability, are emerging as a new trend in AI security governance to prevent data leaks from prompt injection. Concurrently, frequent AI-related attacks (e.g., npm supply chain worms, React2Shell vulnerabilities) necessitate upgraded security protections.