Overview
- DeepSeek, which released preview builds Friday, introduced V4‑Pro and V4‑Flash as open‑source models with a 1 million‑token context window.
- Huawei said its Ascend 950 supernodes fully support V4 and helped train the Flash variant, signaling China’s push to run leading models on domestic chips.
- DeepSeek reports V4‑Pro leads other open models in coding and reasoning and trails top closed systems by roughly three to six months on benchmark tests.
- Published prices undercut major rivals, with V4‑Pro at $0.145 per million input tokens and $3.48 per million output tokens and V4‑Flash at $0.14 and $0.28, though Pro capacity is limited by compute constraints today.
- The debut follows U.S. claims of large‑scale theft of AI technology by Chinese entities and separate distillation accusations against DeepSeek, as reports say Tencent and Alibaba are weighing investment at a valuation above $20 billion.