DeepSeek

    DeepSeek V4 Pro

    Performance close to top closed models

    DeepSeek-V4-Pro significantly improves agent capability over prior generations and reaches leading open-model performance on agentic coding evaluations. It has rich world knowledge and performs competitively with top closed models on math, STEM, and competitive coding reasoning benchmarks.

    Context1M
    Released2026-04
    Relays14 sites
    Top open agentic codingRich world knowledgeMath and STEM reasoningStrong competitive coding

    DeepSeek Official Pricing

    CNY
    Updated: 2026-04Source

    Input

    ¥3/ 1M tokens

    Output

    ¥6/ 1M tokens

    Cache read

    ¥0.025/ 1M tokens

    Relay Comparison

    Compare token, per-request, or per-second pricing by relay channel.