DeepSeek AI released DeepSeek-OCR, a new OCR system using optical 2D mapping and vision-based compression to improve long-text processing for large language models. It delivers high accuracy under heavy compression, outperforms existing models, and supports large-scale digitization, multilingual texts, and efficient AI training. Open-source on GitHub.

Meta introduces new fraud detection features in WhatsApp and Messenger to combat scams targeting elderly users, including screen-sharing alerts and AI-reviewed scam messages. In H1 2025, Meta disrupted 8M fraud accounts and 21K fake support pages. Scammers exploit loneliness and use social engineering across platforms, prompting Meta’s expanded security measures and collaboration with the National Elder Fraud Hotline.

X (formerly Twitter) is testing a new usage-based API pricing model, charging developers based on actual API usage instead of fixed monthly fees. The pay-as-you-go system, currently in a closed trial with selected developers, features separate costs for different API requests and includes a $500 credit for trial users.

Anthropic expands Claude Code to iOS and web, enabling natural language coding via browser or mobile. Users can link GitHub projects, automate code changes, run tests, and manage pull requests in a secure cloud sandbox. The iOS app allows remote code review and task management, positioning Claude Code against GitHub Copilot and Codex amid Anthropic's financial and legal challenges.

OpenAI launched ChatGPT Atlas, an AI-integrated browser with a persistent ChatGPT sidebar, offering web interaction, Agent mode for task automation (Plus/Pro only), and personalized suggestions via memory tracking. Available on MacOS, with Agent mode limited to premium users, Atlas aims to redefine browsing but faces challenges in user adoption and competition.

OpenAI introduces new policy to prevent unauthorized AI-generated celebrity likenesses following backlash from Hollywood and actors’ unions over misuse of Sora 2.

LangChain Inc. raises $125M at $1.25B valuation to advance AI agent development tools.

Google introduces LLM-Evalkit, an open-source framework built on Vertex AI SDKs to streamline and measure prompt engineering for large language models.

Adobe expands enterprise AI with custom generative models via Adobe AI Foundry, enabling businesses to create brand-specific content using Firefly technology.

Meta AI's mobile app usage surged, with daily active users rising to 2.7 million from 775,000 in four weeks. Daily downloads hit 300,000, up from under 200,000. The Vibes video feed launch in September likely drove growth, coinciding with increased engagement. OpenAI’s Sora release may have also boosted interest in Meta AI as an alternative.

Anthropic launched Claude for Life Sciences, a specialized AI model to streamline drug discovery and research workflows, integrating with platforms like Benchling and PubMed. It enables faster data analysis, documentation, and lab protocol interpretation. The company also expanded Claude Code to web and mobile, boosting developer productivity with secure, parallel task execution.

Unitree unveils H2 humanoid robot with 31 joints, enhanced flexibility, and biomimetic design, showcasing advanced mobility and human-like appearance.

Meta introduces parental controls for teen AI interactions, allowing guardians to monitor and restrict chatbot use, aiming to improve safety and rebuild trust amid concerns over minors' engagement. Tools roll out early next year, initially on Instagram in select English-speaking regions.

NVIDIA begins mass production of Blackwell chips in Arizona with TSMC, marking a key shift in its U.S.-based supply chain strategy. The Blackwell architecture, featuring enhanced Transformer and Decompression Engines, powers AI workloads up to 15 petaflops and underpins NVIDIA's latest consumer and enterprise products. TSMC plans further Arizona facilities, including one for its upcoming A16 node. In Texas, NVIDIA will build two "supercomputer manufacturing facilities" expected to begin producti

Mirantis released k0rdent 1.2.0, an open-source "super control plane" for managing Kubernetes across multi-cloud, hybrid, and edge environments. It offers centralized lifecycle management, declarative service deployment, and observability with FinOps capabilities. Key updates include ARM64 support, OpenStack integration, improved Azure flexibility, and enhanced documentation. Built on open standards, k0rdent enables teams to build secure, vendor-neutral internal developer platforms (IDPs) and se

DeepMind collaborates with CFS to apply AI in optimizing nuclear fusion, aiming to control plasma in the Sparc reactor for clean, limitless energy.

Microsoft unveils enhanced Copilot AI features for Windows 11, including Vision support, voice control, file automation, and expanded integrations. Some updates are live, others rolling out via Windows Insider.

Pinterest introduces tools to limit AI-generated content visibility, letting users customize feeds and reduce AI spam across categories like beauty, art, and fashion.

Honor launches Robot Phone, combining AI, robotics, and imaging tech, marking a milestone in its "Project Alpha" and signaling the start of the "smart devices" era. The device, to debut globally at MWC 2026, features a foldable robotic camera and AI-powered multimodal system. Part of Honor’s three-phase AI strategy, it aims to evolve from smartphones to an AI ecosystem, backed by $10B investment and partnerships with Google Cloud, Qualcomm, and Vodafone.

Microsoft launches MAI-Image-1, its first internally developed text-to-image AI generator, designed to create realistic images with advanced lighting and scenery. The model ranks top 10 on LMArena and is set to integrate into Copilot and Bing Image Creator. This follows Microsoft's earlier release of MAI-Voice-1 and MAI-1-preview, signaling its push to diversify AI models and reduce reliance on OpenAI while maintaining partnership.