Models & Labs
Open-weight AI models near top-tier capabilities
TechCrunch AIhigh confidence
Why it matters
- →Open-weight models like GLM-5.2 are nearing the capabilities of leading AI systems, posing new safety challenges.
- →The lack of effective safeguards in open-weight models increases the risk of misuse in cyber and biological domains.
- →Managing the risks of open-weight AI requires balancing accessibility with robust safety measures.
China's Z.ai has developed an open-weight AI model, GLM-5.2, that is rapidly approaching the capabilities of industry leaders like OpenAI and Anthropic. However, a report from SaferAI reveals that GLM-5.2 lacks the safety measures present in its closed-weight counterparts, failing to refuse potentially harmful tasks. This raises concerns about the risks of open-weight models, which can be used without restrictions once downloaded. The debate now focuses on how to manage these risks as open-weight models become more advanced and accessible.
Read originalMore from TechCrunch AI
© TechCrunch AIMarket & Regulationbusiness
Anthropic signs $10B deal with AI cloud startup Volta
Anthropic has made a bold move by securing a $10 billion deal with Volta, a newly established AI cloud startup. This partnership will enable Anthropic to access cloud computing resources from Volta over a six-year period, powered by Nvidia's state-of-the-art Vera Rubin systems. The collaboration also involves Bitdeer, a crypto-mining company, which will contribute to the development of a data center in Norway with a 133 megawatt capacity. This agreement is a strategic step for Anthropic as it seeks to expand its computing capabilities and maintain a competitive edge in the AI industry.
TechCrunch AI
© TechCrunch AIGeneral AIwriting
Wrinkles App Turns Locations into Storytelling Experiences
Wrinkles is a new app that transforms your surroundings into a rich tapestry of stories, acting as an AI-powered audio tour guide. By leveraging your location, it reveals hidden histories and local lore without the need to constantly check your phone. Founded by David Stemler and Ryan Hansan, the app offers a global network of 1.3 million points of interest across 177 countries. Users can interact with these stories, ask questions, and even add personal memories, making the experience both educational and personal. Wrinkles aims to be more than just a travel app, offering insights into everyday locations and fostering a deeper connection with the world around us.
TechCrunch AI
© TechCrunch AIMarket & Regulationbusiness
Nvidia-Led AI Group Shows Early Progress
The newly formed Open Secure AI Alliance, led by Nvidia, is already making strides with the creation of the Shared AI Findings Exchange (SAFE). This working group is actively developing proposals for AI cybersecurity incident reporting and analysis, with the Linux Foundation managing these efforts. While the guidelines are not groundbreaking yet, the rapid collaboration among over 120 companies, including major players like Adobe and Microsoft, signals a strong commitment to open-source AI security. The absence of Google and OpenAI, despite their initial support, adds an intriguing dynamic as the group gains momentum.
TechCrunch AI
More in Models & Labs
© Hugging Face BlogModels & Labsagents
LFM2.5-2.6B: Efficient AI Agents for Local Deployment
Liquid AI's LFM2.5-2.6B model is a breakthrough in deploying AI agents locally, offering performance that rivals models four times its size. It excels in instruction following and tool use, making it a strong contender in these areas. The model's architecture supports rapid inference speeds, even on mobile devices, thanks to its efficient design. With specialized training phases, it handles multi-step agentic tasks effectively. This release empowers developers to integrate advanced AI capabilities into their applications, enhancing accessibility and functionality without the need for extensive resources.
Hugging Face Blog
© Google AI BlogModels & Labsmodels
Google's July 2026 AI Innovations Unveiled
Google's July 2026 AI announcements reveal significant advancements in making AI more practical and integrated into daily life. The introduction of faster Gemini models aims to enhance developer capabilities in scaling production agents and robotics. Meanwhile, the launch of new creative tools in music and video, along with improved AI integration in Search, reflects Google's commitment to expanding AI's role in creative and everyday applications. These developments highlight Google's ongoing efforts to leverage AI for societal benefits, from crisis response to skilled trades training.
Google AI Blog
Models & Labscoding
OpenAI Enhances Education with ChatGPT Work and Codex
OpenAI is expanding the educational capabilities of ChatGPT Work and Codex with new plugins designed for K–12 and college environments. These tools aim to assist teachers and students in learning, teaching, and conducting research more effectively. By integrating AI into the educational process, OpenAI is providing educators with innovative ways to engage students and streamline their workflows. This development marks a step forward in making AI a practical tool in educational settings, potentially transforming how knowledge is imparted and acquired.
OpenAI