Skip to content

Key Models

The following models are among the most commonly used with LLM Compressor: DeepSeek V4 Flash, Qwen3.8-27B, Qwen3.8-2.4T-A95B, Qwen3.8-Flash-Next, Kimi-K3, Gemma 4, Llama 4, Mistral Large 3, HY-V3, GLM-5.3, and GLM-5.3-Flash. Each model page contains quantization examples with tested configurations and recommended parameters.

  • DeepSeek V4 Flash


    DeepSeek V4 Flash with HCA, CSA, and mHC, quantized to FP8 + NVFP4.

    DeepSeek V4 Flash

  • Qwen3.8-27B


    Qwen3.8-27B vision-language model, quantized to INT4, NVFP4, and MXFP4.

    Qwen3.8-27B

  • Qwen3.8-2.4T-A95B


    Qwen3.8's large sparse MoE model, quantized to NVFP4 + FP8 and REAP + NVFP4.

    Qwen3.8-2.4T-A95B

  • Qwen3.8-Flash-Next


    Qwen3.8's Flash sparse MoE model, quantized to NVFP4.

    Qwen3.8-Flash-Next

  • Kimi-K3


    Moonshot AI's Kimi-K3 multimodal model, quantized to NVFP4.

    Kimi-K3

  • Gemma 4


    Google's latest multimodal model.

    Gemma 4

  • Llama 4


    Meta's Llama 4 Scout multimodal model.

    Llama 4

  • Mistral Large 3


    Mistral's 675B parameter model.

    Mistral Large 3

  • HY-V3


    Tencent's HY-V3 sparse MoE model, quantized to NVFP4 + FP8.

    HY-V3

  • GLM-5.3


    THUDM's GLM-5.3 MoE model, quantized to MXFP4 and NVFP4.

    GLM-5.3

  • GLM-5.3-Flash


    THUDM's GLM-5.3-Flash vision-language MoE model, quantized to NVFP4.

    GLM-5.3-Flash