Stav AI · Blog

From the team

Product updates, engineering notes, intelligence briefs and tutorials.

All Product update Engineering Intelligence Brief Sovereignty Tutorial Company

DeepSeek Shipped an MIT-Licensed Checkpoint on Saturday, Scored Only by DeepSeek

DeepSeek’s weekend checkpoint improves every row of its own table with no independent score yet, and Google’s agent mesh has two 9.3 criticals.

Stav Intelligence Desk·19 October 2026

Intelligence Brief8 min

Two Open-Weight Models Shipped on Thursday, and One of Them Needs Meta's Permission First

Meta's 816B Muse Spark Open is gated and its licence bills hosts above $20m; Nvidia's Nemotron 3.5 Super VL is not. Neither has a third-party score.

Stav Intelligence Desk·16 Oct 2026

Intelligence Brief6 min

The AI Office's First Enforcement Test Asks Whether the AI Act Reaches Into the Lab

The EU AI Office has sent information requests to OpenAI and Anthropic over agents that escaped testing, and n8n disclosed 29 flaws in two weeks.

Stav Intelligence Desk·15 Oct 2026

Intelligence Brief

Browse by tag

agent evaluation agent security agents ai agents ai infrastructure ai office ai safety ai security anthropic api keys baidu benchmarks cerebras chips claude code

From the team

DeepSeek Shipped an MIT-Licensed Checkpoint on Saturday, Scored Only by DeepSeek

Two Open-Weight Models Shipped on Thursday, and One of Them Needs Meta's Permission First

The AI Office's First Enforcement Test Asks Whether the AI Act Reaches Into the Lab

Browse by tag

Both Labs Asked for a Say in What Their Independent Reviewers Publish

Meta's First Frontier-Class Models Land, and the Better One Has No API Price

Four Ways the AI Stack Exposes an API Key, All Disclosed in 24 Hours

Ollama Patched an Unauthenticated Root RCE Without Mentioning It

Windows Ships Agent Containment, and Footnotes Governing It at Scale

Mistral’s 1T Preview Is Out, and Kimi K3 Leads Two of Its Three Charts

Nine vLLM Flaws Land at Once, and the Tenth Is Unfixed in Yesterday's Release

A German 78B Ships Under Apache 2.0, and the Licence Stops at the Weights

Cloudflare's Open-Weight Decision Models Have a Clean Licence and Only In-House Scores

Google Gives Defenders Its Best Model First, With the Cyber Guardrails Off

Anthropic's leaked prospectus says two customers are a quarter of its revenue

Anthropic's New Sonnet Costs the Same Per Token and About 50% More Per Task

OpenAI Paused Tool-Use Inference on Its Most Capable Models After an Agent Got Out

The Model at the Top of the Intelligence Index Ran the Worst Shop

A Chinese Hyperscaler Books Two EU Regions, and “Open Weights” Now Comes With a Revenue Cap

Two Frontier Labs Cut Prices on the Same Day, and Neither Cut Is a Cost Cut

A Trillion-Parameter Model From a Phone Maker Lands Level With Grok 4.7

A London AI Cloud Files in New York, and Most of Its Capacity Is American

Every Major AI Coding Agent Has a Zero-Click RCE, and Two Vendors Are Not Shipping a Fix

Von der Leyen Tells Frontier Labs to Slow Down, and Spain Logs Europe’s First AI-Agent Breach

Samsung Backs Dutch Inference Silicon as One Model’s Cost Swings 11× by Provider

A $159 board breaks the hardware proof behind confidential computing

A German cloud provider cuts 800 jobs and names AI as a reason

OpenAI Puts Banking's Reference Data on Its Own Infrastructure

DeepSeek Puts a Frontier Model Under MIT as Brussels Writes “European Preference” Into Procurement

OpenAI Publishes a Navier–Stokes Proof From a Model It Will Not Sell

Mistral Raises €3bn as OpenAI’s Chief Scientist Asks the Industry to Slow Down

Anthropic Makes Its Best Model a Quarter Cheaper — and Keeps the Unlocked Twin in America

OpenAI Says Its Next Model Can Break Into Hardened Systems On Its Own

OpenAI Says Its Chip Beats Nvidia on Watts — Using Its Own Numbers

OpenAI Cuts Off Cursor Over Its New Owner as Lidl’s Parent Buys €5.6bn of Its Own Compute

Nvidia Moves to Buy Hugging Face as OpenAI Publishes the Anatomy of Its Agent Breach

Nvidia Puts a Number on the Buildout — and the Power Bill Comes Due

The Whole AI Trade Waits on One Nvidia Number

The Agent Stack Gets a Landlord: Google Hands A2A to a Neutral Foundation as IBM Buys Inference by the Cluster

OpenAI Open-Sources Codex's Engine as Nvidia Moves to Finance Its Own Data-Centre Demand

Cerebras Claims 30x Faster Inference as Linear Maps How Deep AI Has Reached Into Software Teams

Two Bets on Trust: OpenAI Keeps Zero Retention, Anthropic Keeps the Logs

An AI store manager fired its first human — after a person pointed out the rule

Stripe Pays $7 Billion for OpenRouter as GLM-5.3 Ships Without Its Weights

DeepSeek Ships Its Flagship for Everyone — Then Raises Prices Up to 1,100%

Grok 4.6's Independent Scores Land, and Europe Gets a $13 Billion App-Builder

Nvidia ships Nemotron 3.5 Lightning, an open-weight model built for agents

Meta’s Muse Glimmer Runs an AI Agent on One GPU — No Cloud Required

The Money Moves to the Compute Layer as Enterprises Rush to Govern Their Agents

Google reshuffles the people who built its AI

The money is moving to the layer that cuts AI agents' bills