Stay informed. Keep your time.
8 stories · ~3 min readEstimated reading time for the entire edition.
Super VL has 120 billion parameters; the Bolt models will run locally on AI PCs with RTX Spark and DGX Spark systems. Nvidia says Super VL is faster than competitors on certain agent tasks, a claim based on fine-tuning for those tasks.
Nvidia introduced open-weight Nemotron 3.5 Super VL and two Bolt models for enterprise agents and local computer use.
The order directs federal civilian agencies to stop buying products from DeepSeek and its affiliates and remove them from their systems; it also covers third-party tools built with DeepSeek applications. Agencies may request exceptions within 60 days. The text cites no case of government data reaching DeepSeek.
The US Department of Homeland Security ordered federal civilian agencies to stop buying DeepSeek products and third-party tools using the AI, and remove them from their systems.
Tenstorrent raised $1.5 billion at a $6.2 billion valuation. CEO Jim Keller told Bloomberg they chose to remain independent after demonstrating that Galaxy servers could run DeepSeek models. The new capital will go toward scaling chips for open-weight models and video generation.
Tenstorrent raised new capital to expand its chips for open-weight AI models and video generation.
TorchNeuron lets Hugging Face libraries such as Transformers, Diffusers and TRL run on AWS Trainium with standard PyTorch APIs. According to the post, it supports 89% of the 100 most-used text-generation architectures; dynamic shapes cannot yet be compiled, and asynchronous weight loading is unsupported.
Hugging Face said its TorchNeuron plugin for Trainium, integrated as a native PyTorch device, will enter open beta on October 20, 2026.
MiniMax made M3.1 Flash available through its API, M Plan and MiniMax Code. The company says it competes with leading models on agent, coding and computer-use benchmarks, inference is about three times faster, and agent-task costs are 70% lower than with its previous model. Weights will be released.
MiniMax says M3.1 Flash is available and costs 70% less than its predecessor for agent tasks; model weights will be shared later.
Mercury 3 was introduced as a new diffusion-based large language model. The company says it generates 1,068 tokens per second on NVIDIA GPUs and offers a one-million-token context window; benchmark results are based on the company’s own tests.
Mercury 3 is now available, with speeds of 1,068 tokens per second and a one-million-token context window.
Norforge reported that AI agents requiring little human intervention were used in attacks on more than 100 sites through exposed Cairn instances. Reuters named John Deere, SharkNinja and Cummins among the targets; company statements gave differing details about data access and which sites were affected.
According to Norforge, an attacker used AI agents on the Cairn platform to attack more than 100 sites; Reuters named John Deere and SharkNinja among the targets.
US prosecutors asked that former OpenAI engineer Abdulrahman Alfozan remain detained, alleging he stole company secrets by placing listening devices in its offices. Their court filing cited claims that he bought a one-way ticket to Saudi Arabia and accessed a phone folder called “Saudi Gov Docs.” These are government allegations; the text does not report the case’s outcome.
Prosecutors argued that former engineer Abdulrahman Alfozan, accused of stealing OpenAI secrets, posed a flight risk and asked that he remain in custody.
That’s it for today.The rest of your time is yours.
↑Stories are selected from Hacker News; accessible original articles are summarized and translated with AI. Summaries may contain errors; check the source beneath each story for details. Preparation starts daily at 06:55 Türkiye time (UTC+3), targeting publication before 07:00.
This edition covers articles shared on Hacker News on the date shown above.
Read or listen to a new edition every morning with Kısa for iPhone, save stories for later and keep your streak.
