DeepSeek V4 drops two open-weight models with 1M-context by default, CSA+HCA hybrid attention, and V4-Pro priced at roughly 1/7 Opus 4.7's output cost.
DeepSeek V4's 10× KV-cache compression restructures AI cost economics globally, exposing a structural threat to US lab pricing and strategic positioning.
A Virginia Tech preprint shows model-native skills extracted via sparse autoencoders outperform human-defined skill files in SFT — and produce 41% gains on math via activation-space data selection.
DeepSeek V4-Pro launches with 1.6T parameters, 1M context, and 10× KV cache reduction over V3.2 — multiplying inference concurrency roughly 10× on the same hardware.
Curated AI insights — sent when there's something worth your inbox.