Frontier models, research breakthroughs, deployment stories and the economics of machine intelligence — decoded weekly.
Inside the 90 days of evals, red-teaming and supply contracts that rewired how labs ship models — and why enterprise buyers are already pivoting their 2026 roadmaps.
Pricing cut 28%; new streaming cache APIs collapse latency on long agent chains.
GB300 racks, liquid cooling, and the data-center economics that will define 2027.
DeepSeek-V4 and Qwen-3 close the gap. Where they still fail, and which teams ship them.
Risk tiers, documentation burden, and a field guide for US startups selling into Europe.
Observability, retries, sandboxing — what the quiet winners actually do.
Why SWE-bench and MMLU stopped predicting product wins, and what to measure instead.
CIO survey data on budgets, vendor lock-in fears, and the pilots quietly being killed.
A cluster of April papers argue the old method has new legs on reasoning tasks.
Three unicorns built on the same insight — and what they do differently.