AI Radar

DeepSeekMoE 16B

Available · Language

DeepSeekMoE 16B is an open-weights mixture-of-experts language model from DeepSeek with 16.4B total and about 2.8B activated parameters, released as a Base and a Chat checkpoint with the DeepSeekMoE research paper. Its architecture combines fine-grained expert segmentation with shared experts that are always active. [1]1. Introduction; 3. Model Downloads [2]section 5.1.2, Model Settings Primary source

Timeline of DeepSeekMoE 16B →

Claims and evidence

  • Released The paper that announces the public release of the checkpoint was submitted to arXiv on 2024-01-11, the day of the first repository commit, which already contains the README with the download table. The current README is undated. Primary source[2]submission history, v1; section 1, Public Release [3]commit date and README, 3. Model Downloads
  • Status AvailableWeights are still public and ungated on Hugging Face under the DeepSeek Model License. Primary source[4] [5]
    • Change · Architecture Uses a sparse mixture-of-experts architecture (DeepSeekMoE) instead of the dense transformer of DeepSeek LLM 7B.Compared with DeepSeek LLM Primary source[2]
    • Change · Efficiency DeepSeek reports results comparable with its dense DeepSeek 7B, trained on the same corpus, with about 40% of the computation.Compared with DeepSeek LLM Primary source[1]1. Introduction
    • Open weights YesDeepSeek Model License; commercial use permitted. Primary source[1]3. Model Downloads; 5. License [4]
    • Context window 4096 tokensGiven as sequence length. Primary source[1]3. Model Downloads, Sequence Length
    • Parameters 16.4BTotal parameters of the mixture-of-experts model; the paper gives about 16.4B total and about 2.8B activated parameters. Primary source[1]1. Introduction [2]section 5.1.2, Model Settings
    • Access Open-weights download Primary source[1]3. Model Downloads

    Lineage

    Predecessors

    No known predecessor.

    Successors

    No known successor.

    Based on

    Not derived from another model.

    Variants and derived

    None recorded.

    Siblings

    None recorded.

    All ancestors

    None.

    All descendants

    None.

    Variants

    DeepSeekMoE 16B Base

    Same dates as DeepSeekMoE 16B.

    • Variant Inline variant in this record. Primary source[1]3. Model Downloads [4]

    Also known as: deepseek-moe-16b-base

    DeepSeekMoE 16B Chat

    Same dates as DeepSeekMoE 16B.

    • Variant Inline variant in this record. Primary source[1]3. Model Downloads [5]

    Also known as: deepseek-moe-16b-chat

    Related AI Radar coverage

    AI Radar coverage starts in June 2026; no coverage linked yet.

    All model releases from DeepSeek on AI Radar →