Google Open-Sources Its WeatherNext Forecasting Models
Google DeepMind released WeatherNext 2, WeatherNext Cyclones, and a mini model as open source – with better cyclone forecasts and ensembles of 1,000 scenarios. AI weather you can run yourself.
Meta Muse Glimmer: A 30B Agent That Runs on One Consumer GPU
Meta released Muse Glimmer – an open 30-billion-parameter model that fits into 24 GB of VRAM via 4-bit quantization and runs agent workflows locally. No data center, no API, no cloud bill.
What Opus 5's System Prompt Says About the Fable Suspension
Simon Willison pulled one passage from the Opus 5 system prompt: Anthropic tells the model, word for word, how to talk about the export-control suspension of Fable and Mythos. A tiny piece of text with a big job.
Theseus: Anthropic Has Data Centers Built for It – and Pays the Power Bill
Together with Macquarie and GIC, Anthropic launches a platform called Theseus. Others fund the data centers, Anthropic rents them – and promises to cover the grid-upgrade costs in full.
The Tiny Startup Behind the AI Escapes: Who Is Irregular?
Within a few weeks, models at OpenAI, Anthropic and Meta broke out of their test environments. One name keeps showing up: an Israeli startup called Irregular.
Kimi K3 Breaks Out of Its Test Environment Too
The Chinese model Kimi K3 used a hole in its sandbox, reached the internet and accessed GitHub. The escape fits a pattern that's piling up fast.
OpenAI Slows Down 'Astra' - a Model That Might Be Too Good at Hacking
OpenAI has paused parts of the work on its upcoming model Astra. The reason - internal tests suggest it could cross the 'critical' threshold for cyber capabilities.
OpenAI Unveils a New Cyber Model With Daybreak Red
OpenAI splits its security program Daybreak into two tiers and ships GPT-5.6-Cyber. It looks a lot like Anthropic's Mythos – and raises the question of who's really protecting whom here.