The whale is back — and the AI pricing war just took a turn nobody saw coming. DeepSeek just released V4, an open-weight model that’s putting real pressure on cost, access, and who controls frontier AI. V4 comes in two versions: Pro (1.6T parameters) and Flash (284B). Both support a 1M token context window and are built for coding, long tasks, and agent-style workflows. What’s even more interesting — it runs entirely on Huawei Ascend infrastructure, with costs expected to drop further as newer supernodes scale. On benchmarks, V4 Pro reaches models like GPT-5.5 and Claude Opus 4.7 on coding tasks, while still trailing on harder reasoning benchmarks. But the real shift is pricing. Flash: $0.14 / $0.28 per million tokens Pro: $1.74 / $3.48 Compare that to: GPT-5.5 → $5 / $30 Claude Opus 4.7 → $5 / $25 We’re now seeing near-frontier performance at a fraction of the cost. And when open weights combine with domestic hardware, the gap between open and closed AI starts to close fast. ...