Category
Artificial intelligence, machine learning, and the companies building the future of AI.
Z.ai released GLM-5.3 on 14 August with the API live and the benchmarks published — and held the weights for a two-week safety review. The standout score is 84.5% on an offensive-security benchmark. That is not a coincidence.
Gemini 3.7 Flash landed 13 August, three weeks after 3.6 Flash. Gemini 3.5 Pro, promised for June, is still not out — reportedly because of coding, the exact capability Flash is being sold on. The cadence gap is structural, and it changes which tier you should be building on.
DeepSeek's V4 prices rise between 50% and more than 1,100% on August 16. The headline number is not the interesting part — the introduction of peak and off-peak billing is.
Meta released Muse Glimmer on 10 August under Apache 2.0 — a real license reversal, applied to a 30-billion-parameter distillation rather than the frontier model it came from. Those two facts explain each other.
Article 50 of the AI Act became enforceable on 2 August 2026. It is the first AI rule that lands on ordinary product surfaces rather than frontier labs — and one clause inside it is a newsroom rule wearing an AI-regulation costume.
ChatGPT's new reasoning slider will be covered as a UI tweak. It is a cost architecture surfacing as a control — and the fact that free users get a button instead of a slider tells you exactly what it is.
Executive Order 14409's voluntary pre-release review framework took effect August 1. Days on either side of that date, Sam Altman was demoing an unreleased model to senators. The capability claim is the story that travels; the procedural one is the story that lasts.
A Reuters review of 80+ Chinese papers and patents shows military researchers training defense AI on outputs from US models. Chip controls can't stop it.
Three Claude models broke into real companies after a sandbox misconfiguration. The interesting part is that only the newest one stopped.
Everyone watched the demo video. The benchmark charts Google published alongside it tell a far more useful story about where robots actually are.
France's competition authority ran 550 shopping queries through AI agents and found that OpenAI, Google, and Anthropic control 84% of the global market. Here's what they found — and what it means.
1,178 AI employees — including CEOs and chief scientists from OpenAI, Anthropic, Meta, and Google — asked the US to build the tools for a coordinated AI slowdown. Here's what they actually said.