Runway, an AI startup, unveiled its latest Gen-4 video model, offering improved continuity and control for creating coherent narratives. The model enables users to generate consistent characters and objects across multiple shots using a single reference image. Currently available to paying customers, Gen-4 allows angle-specific outputs based on user descriptions. This follows the launch of Runway's third-generation Alpha model less than a year ago, which extended video lengths but faced controve

Google's Gemini 2.5 Pro (experimental version) AI model is now free for public use, following limited access for Gemini Advanced subscribers. It features a step-by-step "thinking" process, making it the most advanced model yet and outperforming competitors in LLM benchmarks. Google plans to integrate it into its mobile app and offers enhanced tools for real-time coding. Free access aims to gather user feedback for further improvements.

Amazon's AI-driven Alexa Plus service is now in an early access phase for select customers. Not all features from Amazon's recent launch are included, such as Grubhub ordering or Kids Plus functionalities. Currently compatible with Echo Show 8, 10, 15, and 21, Alexa Plus can order Uber rides, identify objects, draft emails, and search for products. Document summarization is available, but file deletion is still under development. Priced at $19.99 per month, it's free for Prime members, with new

OpenAI's ChatGPT added an image generation feature, allowing AI art in styles like Studio Ghibli, boosting user growth by a million in one hour. However, the surge strains GPU resources, prompting calls for responsible usage. OpenAI is optimizing performance to handle demand while ensuring stability.

Runway AI launches Gen-4, its latest model for creating videos from natural language prompts. Built on neural networks, Gen-4 ensures visual consistency across frames and allows editing objects, angles, and lighting. It also functions as an image editing tool, generating multiple variations. Initially supporting 5-10 second clips, Runway plans future updates for longer content. This release comes as OpenAI temporarily disables Sora's video feature due to high traffic.

OpenAI plans to release a new open-source AI language model with strong reasoning capabilities. CEO Sam Altman announced this shift, acknowledging past hesitations on open-source. The move follows discussions around the DeepSeek incident and aims to drive industry innovation and expand OpenAI's influence in AI technology.

Microsoft expands AI features to Copilot Plus PCs with Intel and AMD chips, including real-time captioning in multiple languages and Cocreator in Paint for AI-generated images. Previously limited to Qualcomm-based PCs, updates also enhance voice access and introduce screenshot search via Recall. Microsoft plans to bring more features, like 27-language translation into Simplified Chinese, to Intel and AMD devices.

OpenAI has raised $40 billion in a funding round led by SoftBank, valuing the company at $3 trillion. This is the largest financing for a private tech firm. OpenAI will initially receive $10 billion, with $7.5 billion from SoftBank. The remaining $30 billion is contingent on becoming a for-profit entity. Despite explosive user growth and plans for new models, OpenAI continues to consume significant capital and may not achieve positive cash flow until 2029. The funding supports its mission to dev

A new report by aitools.xyz highlights DeepSeek, a fast-growing Chinese AI startup, surpassing ChatGPT in February 2025 with 524.7 million new website visits. Despite ranking third overall, DeepSeek's market share rose from 2.34% to 6.58%, and it captured 12.12% of the chatbot market, securing second place. Global AI tools saw 12.05 billion visits in February, reflecting widespread adoption. DeepSeek's robust growth positions it as a key player in the expanding AI market.

Zhipu launched AutoGLM "Contemplation," the world's first AI agent integrating deep research and operational capabilities. Built on models like GLM-4 and GLM-Z1, it enables machines to think and act simultaneously. Key features include deep thinking, environmental perception, and tool usage. Powered by the Contemplation Model, it uses reinforcement learning for long-term reasoning and task execution. Available on Zhipu platforms, it marks a major advancement in China's AI technology.

CloudMinds, a prominent humanoid robotics company, reportedly shut down, leaving its headquarters empty and employees struggling with unpaid salaries. Since February last year, the company reduced wages, halted payments by March-April, and ceased social security contributions by June. Legal issues and supplier debts further highlight its financial struggles. Despite previous strategic agreements and significant VC backing from firms like Alibaba and SoftBank, CloudMinds failed to meet market exp

Google DeepMind introduces TxGemma, an open-source AI suite to accelerate drug discovery and clinical trial predictions. Built on the Gemma family, it uses large language models to enhance therapeutic property predictions, offering tools for target identification and outcome evaluation. With 7 million training examples and three sizes (2B, 9B, 27B parameters), TxGemma excels in classification, regression, and generation tasks. The 27B Predict model performs well in 64 of 66 key tasks. TxGemma-Ch

Aravind Srinivas, CEO of Perplexity, addressed user concerns on Reddit about financial troubles and cost-cutting measures. He clarified that the Auto mode was introduced to simplify the user experience due to clutter in AI products. Srinivas reassured users that the company has sufficient funds, growing revenue, and no plans to go public before 2028.

Apple is developing a new health app version with an AI coaching feature for personalized health advice. Set to launch in spring/summer 2026 with iOS 19.4, the coach will use data from medical devices and focus on areas like food tracking. It's being trained with physician-provided data and may include doctor-recorded videos. The service is tentatively named Health+.

Elon Musk's xAI acquires social media platform X (formerly Twitter) in an all-stock deal valuing xAI at $80 billion and X at $33 billion, including $12 billion debt. This integration leverages X's data for xAI's Grok tool, expanding its reach beyond X to platforms like Telegram. Despite a temporary dip in X's valuation since 2022, it has rebounded alongside xAI's rising worth. Musk emphasizes X as a "digital town square," while closely linking the two businesses operationally.

Huawei open-sourced its ModelEngine AI toolchain at the 2025 China Partner Conference. The toolchain addresses data engineering, model training, and deployment challenges in AI implementation. It covers data, model, and application enablement, offering capabilities like data processing, model fine-tuning, and AI app development. Hosted on multiple platforms, it aims to accelerate industry-specific AI adoption and foster an open-source ecosystem.

Qwen2.5-Omni-7B, the first open-sourced end-to-end multimodal model, processes text, images, audio, and video while generating real-time responses and speech. It sets industry benchmarks in multimodal tasks, surpassing models like Google's Gemini-1.5-Pro. Leveraging Thinker-Talker architecture and TMRoPE, it achieves human-level performance in speech generation and supports industrial applications with its compact 7-billion-parameter design. Available on ModelScope and Hugging Face, Qwen2.5-Omni

Anthropic PBC released two papers revealing insights into LLM information processing. The research highlights Claude's multistep reasoning, language-agnostic components, and planning capabilities. It also shows Claude can adapt its strategies and transfer knowledge across domains. However, the model's explanations of its reasoning don't always match its actual processes, raising reliability concerns. Advances in observation methods are needed to better understand complex LLM behaviors.

OpenAI now supports Anthropic's MCP protocol, enabling large language models to interact with external systems. Integrated into OpenAI’s Agents SDK, MCP will also be added to ChatGPT’s desktop client and Responses API. This open-source technology simplifies connecting LLMs to external data sources, enhancing practical applications like generating shopping recommendations or executing scripts. The latest MCP version includes JSON-RPC batch processing, simplified notifications, and OAuth 2.1 secur

Telegram integrated the Grok AI chatbot into its platform, enhancing user interaction with diverse information and conversational capabilities. This feature is exclusive to Telegram Premium subscribers, reflecting a strategy to boost revenue. The collaboration with Elon Musk's Grok AI underscores shared values on privacy and free speech.