History
DeepSeek presents V4-Pro and V4-Flash
Preview versions of DeepSeek V4-Pro and V4-Flash appeared on 24 April 2026 with a one-million-token context window.

With long documents, you want to know how much information a model can take into one task. On 24 April 2026, The Next Web described preview versions of DeepSeek V4-Pro and V4-Flash. The source mentioned open weights and a one-million-token context window. Tokens are the pieces into which a model divides text.

The news therefore offered two model variants and room for a large input. It was still a preview release. The size of the window does not mean an answer uses every detail correctly. Nor does this event mean a later stable version was available on that day.

For a small business, a public collection of manuals makes a suitable exercise. Ask questions whose answers appear in different places. Always request the passage used and check it. Include a question the documents do not answer.
Then decide whether the information found helps an employee finish a task. A long input is a possibility, not a goal in itself. Want to investigate how to make document knowledge useful inside a tool? Discuss your question with Denkbaar.