
Meta Ships Muse Glimmer, an Open 30B Agent Model
Meta's new 30B Muse Glimmer runs agentic AI tasks on a single consumer GPU under Apache 2.0. The real EU story is who owns the audit trail when it goes wrong.

Meta's new 30B Muse Glimmer runs agentic AI tasks on a single consumer GPU under Apache 2.0. The real EU story is who owns the audit trail when it goes wrong.

Nine of the top 10 text-to-video AI models on Artificial Analysis are built by Chinese labs. What that ranking means for an EU or UK buyer choosing an AI video vendor.

Google DeepMind open-sourced WeatherNext, a peer-reviewed AI model that gives 3-day cyclone forecasts the accuracy 2-day forecasts had before - a full extra day of warning, free to download on GitHub.

OpenAI says ChatGPT has crossed 1 billion weekly users and is making GPT-5.6 Luna the free default with unlimited text chat, as Bloomberg reports a 2027 hardware device with Jony Ive's LoveFrom.

Anthropic rewrote and retrained Fable 5's biology safety classifier, cutting false-positive refusals on benign biology and medical queries by about 85%, while virology and toxicology stay hard-blocked.

Stanford and Arc Institute's Evo AI designed working bacteriophage genomes, first previewed in 2025 and now peer-reviewed in Science. The real gap is DNA-synthesis screening, not the model.

Shopify's Q2 2026 earnings show AI-driven storefront traffic and orders tripling year over year, without traditional search shrinking, a rare additive signal for e-commerce.

Cloudflare open-sourced Cloudflare OS on August 5, 2026: 4,000 staff-built AI apps in 30 days, and the identity-based access model EU and UK IT buyers will adopt next.

Meta launched Muse Code on Aug 5, 2026, with a $0.30/1M token tier that trades your source code for training rights. Here is who should never take it.

Mistral AI's new Shieldstral classifier runs on one 16GB GPU, letting EU platforms moderate content without ever sending user data to a foreign API.

Demis Hassabis moves to chairman and Alphabet chief scientist as Jeff Dean, Sanjay Ghemawat and Oriol Vinyals leave to found Discovery Loop. Alphabet stock fell nearly 4 percent, and Koray Kavukcuoglu now runs Gemini day to day.

Microsoft Research's open-source Orchard framework runs agent-training sandboxes about 10 times cheaper than managed rivals, and a model trained with it reaches 69.7 percent on SWE-bench Verified using just 3 billion active parameters.
Page 3 / 8
One considered note on infrastructure, governance, and measurement, most mornings. No theory.