DeepSeek previews V4-Pro (1.6T) and V4-Flash with 1M context

DeepSeek released a preview of V4 in two versions: V4-Pro at 1.6T parameters and V4-Flash at 284B, both supporting 1M-token context. Open-source, MIT-licensed, with day-one vLLM support and pricing significantly below US frontier labs. Strong agentic and coding benchmarks rival closed-source frontier systems.