Tencent open-sources HunYuan-MT, a new series of translation-optimized language models, outperforming Google Translate and other leading AI models in benchmark tests.

Chinese researchers develop world’s first full-spectrum 6G chip, enabling mobile speeds over 100 Gbps. The 11×1.7 mm chip covers 0.5 GHz–115 GHz, integrating key components on thin-film lithium niobate. It uses photonic-assisted signal generation for fast, stable microwave-to-terahertz transmission. This breakthrough, published in *Nature*, advances reconfigurable, ultra-wideband wireless networks, potentially enabling 6G commercialization by 2030.

LayerX, a Japanese AI SaaS platform, raised $100M in Series B funding led by TCV to scale back-office automation. With AI-driven products like Bakuraku, Alterna, and Ai Workforce, it targets financial and HR workflow efficiency amid Japan’s digitalization challenges.

OpenAI plans to build a 1GW data center in India as part of SoftBank-backed Stargate project; supports sovereign AI infrastructure via "OpenAI for Countries" and expands local presence with ChatGPT Go, facing competition and legal challenges.

Mt. today released and open-sourced LongCat-Flash-Chat, a 560B-parameter MoE model with dynamic activation (18.6B–31.3B), efficient inference (100 tokens/sec on H800), and strong performance across agentic tasks, programming, and instruction-following.

NVIDIA's Q2 revenue saw 40% from two unnamed clients, "Customer A" (23%) and "Customer B" (16%), amid a record $46.7B quarter driven by AI demand. While concentration risk exists, analysts note these clients' strong financials and expected data center investments.

Meta explores integrating Google's Gemini and OpenAI's models into its apps temporarily while advancing its own Llama 5 AI. It adopts a hybrid strategy, partnering with Midjourney for enhanced image/video tools and hiring top talent amid internal restructuring. Despite challenges with Llama 4 and chatbot controversies, Meta invests heavily in AI R&D, raising capex to $66B–$72B annually.

Maisa AI secures $25M seed funding to scale its AI agent platform, enabling businesses to deploy AI without coding or training data.

OpenAI launched its Realtime API, featuring the "gpt-realtime" Speech-to-Speech model for voice assistants in customer service, education, and productivity. It offers faster responses, natural voices, supports non-speech signals, language switching, customizable tones, and new voices. Accuracy improved significantly in benchmark tests. The API enhances tool integration, supports image inputs, lets developers control image access, and adds token limits and conversation condensing to manage costs.

xAI launched Grok Code Fast 1, a cost-effective code generation model excelling in TypeScript, Python, Java, Rust, C++, and Go. Priced at $0.20/million input tokens, it offers free access on platforms like GitHub Copilot and Cursor. The model features a new architecture optimized for real-world coding tasks, from debugging to project development. Major AI firms like OpenAI and Microsoft also expand AI code tools, with Microsoft reporting 20%-30% AI-generated code usage. Meanwhile, xAI sued Apple

Anthropic mandates user opt-in by Sept 28 for data use in AI training, marking a shift from prior 30-day auto-delete policies. Affected users (Claude Free/Pro/Max) can opt out, while enterprise clients remain exempt. The move reflects broader industry trends, raising concerns over consent clarity and data retention practices amid growing regulatory scrutiny.

Microsoft introduces OOBE updates for Windows 11, improving security and stability before first login, available to business and education users from September 2025.

Google updates Vids with AI avatars, image-to-video tools via Veo 3, transcript trimming, and format flexibility, enhancing team video creation and editing.

Anthropic PBC launched "Claude for Chrome," an experimental browser extension allowing its AI to control Google Chrome, aiming to enhance AI integration with web interfaces while addressing security risks like prompt injection and unauthorized access.

Google launched Gemini AI on Google Distributed Cloud, enabling secure on-premises deployment for enterprises and governments with strict data compliance needs.

Hyland unveils Content Innovation Cloud updates with Enterprise Context Engine and Agent Mesh, enabling AI-driven automation via unified data and task-specific agents across industries.

WhatsApp rolls out "Writing Assistant," an AI-powered tool to help users refine messages with customizable tones. Initially available in English in the U.S. and select countries, it offers alternative versions of texts—more formal, supportive, concise, or grammatically correct—without revealing AI involvement to recipients. Integrated into chats, it activates via a pencil icon and works securely using private computation techniques, ensuring WhatsApp/Meta cannot access or store messages. Process

Vercel, a front-end development startup, is raising several hundred million dollars in new funding led by Accel, potentially valuing the company at $9 billion. Known for Next.js, Vercel offers cloud services for rapid web updates and AI tools like v0, AI Gateway, and low-cost GPU access.

OpenAI and Anthropic collaborated on joint AI safety testing, sharing restricted model access to identify evaluation blind spots amid rising AI risks. The study revealed differences in hallucination handling and "assistant gratification" tendencies, highlighting urgent safety challenges as AI systems grow more powerful and widely used.

Attio raises $52M to build AI-native CRM platform, enabling real-time data integration, automation, and customizable workflows for enterprise growth.