Mira Murati's Thinking Machines Lab has officially debuted Inkling, a massive 975-billion-parameter open-weights multimodal Mixture-of-Experts (MoE) model. Featuring a 1-million-token context window and controllable reasoning effort levels, the model natively processes text, images, and audio. Positioned as a highly customizable base model for fine-tuning, Inkling has launched with immediate day-zero support across major inference frameworks.
Sources: Ben's Bites, Digg AI, Hugging Face Blog, HuggingNews, Keep Bookmarks, Latent Space, OpenAI Blog, Reddit AI, Simon Willison, The Neuron, The Rundown AI
Walden Robotics, a spinoff of the Toyota Research Institute, has officially launched from stealth with $300 million in funding and a $1.1 billion valuation. Backed by Toyota, the company has already deployed its LBM-powered industrial humanoid robots into active production environments. This transition marks a major milestone in moving advanced AI-driven robotics from research labs to practical industrial applications.
As tech giants like OpenAI and Oracle push forward with controversial data center projects despite local opposition, the Trump administration has signaled support for overriding state objections to accelerate AI infrastructure. Meanwhile, the political battleground is intensifying with AI-linked super PACs spending millions to lobby on safety regulations, even as tech companies like Apple, Google, and Chinese labs work on smaller models to reduce reliance on massive data centers.
New York Governor Kathy Hochul has signed an executive order pausing permits for new data centers exceeding 50 megawatts for one year to establish standardized frameworks for future facilities. Former President Donald Trump has urged the governor to reverse the decision, warning that the temporary halt could hinder state revenue and push critical AI development to China.
Linux creator Linus Torvalds has strongly defended the integration of AI and Large Language Models into the Linux kernel project, viewing them as indispensable tools for developers. Pushing back against internal opposition, Torvalds emphasized that technical merit guides project decisions and stated that those who disagree with this direction are free to fork the project.
OpenAI is rumored to be developing a portable, screenless smart speaker with advanced AI capabilities, cameras, and sensors, signaling its ambition to integrate AI into physical environments beyond chat interfaces.
AI safety startup Anthropic is reportedly exploring an initial public offering, indicating a significant financial milestone for one of the leading frontier AI developers.
TSMC announced a further $100 billion investment in Arizona for advanced 2-nanometer and below semiconductor facilities, increasing its total U.S. commitment to $265 billion. This significant expansion is primarily fueled by the robust and projected long-term demand for AI-driven chips, which accounted for a large portion of its recent revenue growth.
Chinese AI startup Moonshot has launched Kimi K3, a massive 2-3 trillion parameter open-weight language model designed to challenge Anthropic's Claude Opus 4.8. Alongside this major product release, the company is reportedly raising a new round of capital that would value the startup at approximately $31.5 billion. This dual milestone highlights both rapid technological advancement and strong investor confidence in the Chinese AI sector.
Apple's AI service, Apple Intelligence, has received regulatory approval in China. Alibaba's Qwen model and Baidu's technology will power the localized rollout across Apple's operating systems, leading to a significant jump in Alibaba and Baidu shares.
OpenAI has integrated its new GPT-5.6 model into Codex, offering double the token efficiency and half the pricing alongside advanced features like parallel processing and real-time visualization. This update comes as OpenAI's developer platforms, Codex and ChatGPT Work, surpassed 9 million active users following a 2.5-fold weekly surge in agentic product usage. To accommodate this rapid growth and maintain system stability, OpenAI has successfully restored full weekly usage quotas and lifted temporary session limits.
Chinese AI startup DeepSeek is preparing for a secondary funding round that could value the company at approximately $71 billion. This valuation is backed by strong financial performance, with annualized revenue approaching $500 million and gross margins of 70-80% on its V4 API. To secure further capital for expansion, the company is also planning a public listing on the Shanghai STAR Market.
The Los Angeles Police Department and dozens of other agencies across the US have terminated contracts with AI surveillance provider Flock Safety due to severe privacy, data misuse, and civil liberties concerns. Additionally, a new study has alleged racial bias in the placement of these automated license plate readers, while courts and civil rights organizations like the ACLU grapple with the Fourth Amendment implications and advocate for community oversight.
A German research consortium, funded by the Federal Ministry for Economic Affairs and Energy, has released Soofi S, an open-source 30B language model trained on Deutsche Telekom's Industrial AI Cloud. While the model adheres to the Open Source AI Definition 1.0, it has faced criticism regarding potential overtraining, which developers defend as a benefit of its Mixture-of-Experts architecture. The project represents a major step for European sovereign AI infrastructure, despite navigating complex regional data licensing challenges.
A security exploit in Nvidia's Falcon security processor may allow the CMP 170HX, a GPU previously limited to crypto mining, to be reverted to its full A100 80GB compute and memory capabilities. This discovery could significantly reduce the cost of acquiring high-performance AI/ML hardware.