MiniMax-H3 Pruned Model INT8 Quantized Versions Released

INT8 and INT8 ConvRot quantized versions of the MiniMax-H3 Pruned Ref-Delta Fused r1024 model have been released on Hugging Face, optimized for ComfyUI. The FC2 layer is kept in BF16 to balance compression and performance.

2026-08-22 ~ 2026-08-22 · 2 related posts

• MiniMax-H3 INT8 Release: Why Keep FC2 in BF16 — marres · 2026-08-22
• MiniMax-H3 Pruned Model Released: INT8 and ConvRot ComfyUI Versions — marres · 2026-08-22