DeepSeek Unveils V4 AI Models, Claims Near-Frontier Performance

DeepSeek, a Chinese AI lab, has just unveiled two preview versions of its new large language model, DeepSeek V4, which is a big step up from the V3.2 and R1 versions. The V4 Flash and V4 Pro are both designed as mixture-of-experts models, capable of managing huge documents or code with 1-million-token context windows. The Pro model is particularly impressive, featuring 1.6 trillion parameters, making it the largest open-weight model out there, surpassing competitors like Moonshot AI's Kimi K 2.6. While the V4 Flash has 284 billion parameters, both models are said to perform better than V3.2. They also offer competitive pricing, with the V4 Flash costing $0.14 per million input tokens.