Open Source

Liquid AI Releases LFM2-VL: Open-Weight Vision-Language Models for On-Device AI Liquid AI has released LFM2-VL, a new family of open-weight vision-language foundation models created for low-latency, on-device deployment. The new models are designed to run efficiently on smartphones, laptops, wearables, and other embedded systems without needing to rely on cloud infrastructure. LFM2-VL is available in two versions: LFM2-VL-450M for highly resource-constrained devices and the more powerful LFM2-VL-1.6B. These models can process both text and images, offering up to twice the inference speed on a GPU compared to similar existing models. The models are available on Hugging Face under a license based on Apache 2.0, allowing for free academic and research use, as well as commercial use for smaller companies. ...

Cyware Launches Open-Source AI Server to Secure Threat Intelligence At the Black Hat USA 2025 conference, cybersecurity firm Cyware announced the launch of its open-source Model Context Protocol (MCP) Server. This new tool is designed to securely connect large language models (LLMs) with threat intelligence and automation systems. The MCP Server allows security teams to use natural language inputs to initiate real-world actions and investigations while maintaining full context and control. The move reflects a broader trend of viewing AI as a practical tool for improving security operations by reducing investigation times and improving data correlation. By open-sourcing the server, Cyware aims to foster collaboration and interoperability within the cybersecurity community. ...

Medallion Raises $43M to Streamline Healthcare Credentialing with AI Platform Medallion, a healthcare technology company, has secured $43 million in new funding for its AI-powered platform, CredAlliance. The platform is designed to streamline the complex healthcare credentialing process for both insurers and clinicians. It manages the entire provider lifecycle—including credentialing, enrollment, onboarding, and compliance—aiming to significantly reduce administrative waste and combat clinician burnout. CredAlliance acts as a unified system for provider data management, enabling customers to efficiently verify credentials and maintain up-to-date network information. This latest investment, led by Acrew Capital, elevates Medallion’s total funding to $130 million and will fuel the expansion of the platform’s capabilities. ...

Beijing Hosts Inaugural World Humanoid Robot Games The first-ever World Humanoid Robot Games have commenced in Beijing, drawing 280 teams from 16 countries. This three-day event showcases the latest advancements in humanoid robot technology through a series of demanding competitions. Robots are competing in 26 different events, including athletic challenges like running and football, alongside skill-based tasks such as cleaning and sorting pharmaceuticals. The games are designed to test the robots’ intelligent decision-making and collaborative movement capabilities in extreme scenarios. This event, following the 2025 World Robot Conference, aims to accelerate the real-world application of humanoid robots in sectors like manufacturing, healthcare, and domestic services. ...

Google Launches Gemma 3 270M: A Compact AI Model for Specialized Tasks Google has introduced Gemma 3 270M, a new and compact AI model featuring 270 million parameters. This model is engineered for efficient, task-specific fine-tuning and offers robust instruction-following capabilities directly out of the box. Its minimal footprint makes it ideal for on-device and research applications, enabling developers to build and deploy multiple custom models for tasks like text classification and data extraction without requiring significant computational overhead. Google is providing both pretrained and instruction-tuned versions of the model through platforms such as Hugging Face, Ollama, and Kaggle. ...

OpenAI Releases GPT-OSS, Its First Open-Source Large Language Models In a significant move for the AI and open-source software communities, OpenAI has released GPT-OSS, its first-ever family of open-source large language models. Licensed under the permissive Apache 2.0, this release introduces two powerful models: gpt-oss-120b and gpt-oss-20b. Both are engineered for efficient inference and strong reasoning, leveraging a Mixture-of-Experts (MoE) architecture. The larger gpt-oss-120b model rivals OpenAI’s proprietary o4-mini and operates on a single 80GB GPU. Meanwhile, the compact gpt-oss-20b model, comparable to o3-mini, is optimized for on-device AI applications, requiring as little as 16GB of memory. Complementing the models, OpenAI also launched ‘Harmony,’ an open-source project defining a new prompt template format. ...

How Google AI’s New Method Reduces LLM Training Data by 10,000x Google Research has unveiled a groundbreaking method for fine-tuning large language models (LLMs) that reduces required training data by up to 10,000 times. This innovative approach utilizes active learning to focus expert labeling on the most informative examples, especially “boundary cases” where model uncertainty is highest. In experiments with Gemini Nano models, this technique matched or surpassed the quality of models trained on 100,000 random labels with as few as 250 to 450 targeted examples. This development promises to make AI model development significantly leaner, more agile, and more cost-effective. ...

OpenAI Launches GPT-5, Its Most Advanced AI Model Yet OpenAI has officially launched GPT-5, its latest and most powerful AI system. The company touts the new generative AI model as a major leap in artificial intelligence, showcasing state-of-the-art capabilities in coding, mathematics, creative writing, and health. GPT-5 operates as a unified system, engineered to provide rapid answers for simple queries and engage in deeper, longer ’thinking’ for more complex problems. The model is now available to all users, with premium access and features reserved for Plus and Pro subscribers. OpenAI CEO Sam Altman highlighted the model’s evolution, comparing GPT-3 to a high school student and GPT-4 to a college student, while describing GPT-5 as a ‘PhD-level expert.’ This landmark release is a significant move for OpenAI, poised to solidify its leadership in the competitive generative AI industry. ...

Zoox Gets Federal Green Light for Driverless Robotaxi Deployment Amazon’s subsidiary, Zoox, has received a crucial federal exemption from the National Highway Traffic Safety Administration (NHTSA) to deploy its purpose-built, driverless robotaxis on public roads. This is a landmark decision, making Zoox the first U.S. company to gain such an exemption under the expanded Automated Vehicle Exemption Program (AVEP). The custom-designed vehicles notably lack traditional controls like steering wheels or pedals, featuring a carriage-style interior with inward-facing seats for four passengers. As a condition of the approval, Zoox must remove any claims that its vehicles comply with all Federal Motor Vehicle Safety Standards. This regulatory milestone clears the path for Zoox to scale production at its California facility, as the NHTSA has also closed its investigation into the company’s self-certification processes. ...

EU AI Act’s New GPAI Rules Take Effect, Mandating Data Transparency In a significant step for AI governance, new provisions from the EU’s Artificial Intelligence (AI) Act for general-purpose AI (GPAI) models became effective on August 2, 2025. Providers of new GPAI models must now maintain current technical documentation and publish detailed summaries of their training data. More stringent requirements are set for GPAI models identified as posing a “systemic risk.” To streamline compliance, the European Commission’s AI Office has issued a mandatory template for training data disclosure. While the complete AI Act will apply from August 2, 2026, these specific rules for GPAI models are now enforceable. A voluntary Code of Practice has also been released to guide companies in meeting their legal obligations. ...