DeepSeek V4 Preview and the 1M-context open model race

DeepSeek's V4 Preview is another signal that Chinese AI labs are competing hard on context length, active parameters, and cost.

Source

DeepSeek V4 Preview Release

Why I saved it

DeepSeek is important to track because it changed how many developers think about open models and cost-performance.

The V4 Preview release keeps that pattern going: large context, open-source positioning, and separate Pro and Flash variants.

My notes

  • DeepSeek-V4 Preview is presented around cost-effective 1M context.
  • DeepSeek-V4-Pro is listed as 1.6T total parameters with 49B active.
  • DeepSeek-V4-Flash is listed as 284B total parameters with 13B active.
  • The API supports developer workflows through familiar interfaces.

What I want to remember

Long context is becoming normal, but long context alone is not enough. I still want to see retrieval quality, tool use, latency, cost, and how stable the model is over a full agent loop.