Submit

deepseek-v4 by DeepSeek

4 tracked versions of the deepseek-v4 family. Timeline ordered newest → oldest. Flagship version marked.

Version timeline

  1. unknown
    Version
    —
    Context
    1049K
    Parameters
    —
    License
    —

    DeepSeek V4 Flash is a previous-generation text MoE model. DeepSeek retired the V4 Flash API on September 10, 2026. The legacy deepseek-v4-flash alias temporarily invokes V4.1 Flash, so it no longer identifies the original V4 model. The historical model page and weights reference are retained.

  2. DeepSeek V4 Flash 0731FlagshipOpen-weight
    unknown
    Version
    0731
    Context
    1311K
    Parameters
    —
    License
    —

    DeepSeek V4 Flash is a previous-generation text MoE model. DeepSeek retired the V4 Flash API on September 10, 2026. The legacy deepseek-v4-flash alias temporarily invokes V4.1 Flash, so it no longer identifies the original V4 model. The historical model page and weights reference are retained.

  3. DeepSeek V4 ProOpen-weight
    unknown
    Version
    —
    Context
    1000K
    Parameters
    —
    License
    —

    DeepSeek V4 Pro is DeepSeek’s text reasoning model for coding and agent workflows. The current deepseek-v4-pro API serves the August 13 GA revision with a one-million-token context and up to 384K output tokens. DeepSeek confirms that service continues after September 14, 2026.

  4. unknown
    Version
    0813
    Context
    1000K
    Parameters
    —
    License
    —

    DeepSeek V4 Pro is DeepSeek’s text reasoning model for coding and agent workflows. The current deepseek-v4-pro API serves the August 13 GA revision with a one-million-token context and up to 384K output tokens. DeepSeek confirms that service continues after September 14, 2026.

About DeepSeek

See every provider hosting DeepSeek models at /provider/deepseek.