Mid-Year Review: The Year Privacy Became an Infrastructure Problem
Eight hundred data centers under construction. 101 gigawatts of announced private gas generation. Two-thirds of new AI facilities sited in water-stres...
openWeights
5 articles on openWeights.
Frontier vs Open on Security Work — And What That Means When You Hand an Agent Your Shell
Semgrep benchmarked ten models on finding real IDOR vulnerabilities in open-source code. Claude Opus 5 led at 65.6% F1; GPT-5.6 Luna hit 48.0% at a fi...
Which Open-Weights Model Can You Actually Run? An August 2026 Field Guide for Private AI Assistants
Three near-frontier open-weights releases landed inside one week in August: Qwen3.8-Max on the 12th, DeepSeek V4 Pro on the 13th, GLM-5.3 on the 14th....
The Open-Weights Gap Closed in Two Years — Why That's a Privacy Story
At the end of 2023 the best closed model scored around 88% on MMLU and the best open one about 70.5% — a 17.5-point gap. By early 2026 that gap on kno...
When the Best Model in the World Vanishes in Three Days: The Case for Local AI
Anthropic's Fable 5 launched on June 9 and was pulled days later after a Commerce Department directive. Whatever you think of the policy, the lesson i...