DeepSeek Updates V4-Flash and Teases Upcoming V4-Pro Release
DeepSeek has updated its DeepSeek-V4-Flash model, which reportedly outperforms Z.ai's GLM 5.2 at the same cost as the previous version, and announced that the official release of DeepSeek-V4-Pro will follow soon. As DeepSeek continues to drive innovation in the open-weights LLM space, these updates provide developers with highly efficient, cost-effective models for coding and agentic workflows. DeepSeek-V4-Flash is a Mixture-of-Experts (MoE) model with 284B total parameters (13B active) and a 1M context window, while the upcoming V4-Pro is a larger 1.6T parameter model (49B active).
## BACKGROUND
DeepSeek is known for releasing highly efficient Mixture-of-Experts (MoE) models that rival proprietary alternatives at a fraction of the training and inference cost. Mixture-of-Experts is an architecture that activates only a subset of parameters for each token, allowing large models to run faster and cheaper.