On February 11th, many users discovered that the DeepSeek web version and mobile application were simultaneously updated. The most notable breakthrough in this upgrade is the context window increased to 1M (1 million) Tokens, marking its entry into the global top tier in large-scale document processing.

DeepSeek

From 128K to 1M: A Leap in Processing Power

Compared to the DeepSeek V3.1 released in August of last year (with a context length of 128K), this update has improved processing power by nearly eight times.

  • Test Results: According to user tests, DeepSeek can now "absorb" an entire long novel (such as "Jane Eyre" with over 240,000 tokens) in one go and accurately identify and analyze the document content.

  • Application Scenarios: This means programmers can upload entire project code repositories, scholars can compile millions of words of thesis materials at once, and office workers will find it much smoother to handle extremely long meeting minutes or legal contracts.

The Flagship Model DeepSeek V4 Is Still "Gathering Momentum"

Although this update significantly enhanced long-text capabilities, industry insiders suggest that this may only be a minor iteration of the V3 series. The real highlight—the next-generation flagship model of DeepSeek (expected to be V4)—is currently in the final stages of development.

According to insiders, due to the model's scale reaching the "trillion-parameter level," the increased training complexity has slightly delayed the release schedule.

This update has attracted widespread attention in the tech community. Netizens have expressed that the million-token context is a "major breakthrough" for programming and big data analysis. At the same time, models such as QwenLong are also making efforts in the field of long-text reasoning, and the "long-text competition" among domestic AI is becoming increasingly intense.