MiniMax-M3-uncensored · 428B MoE

Published 2026-07-091 min read

Study refusal behavior in a 428B multimodal MoE while preserving its reasoning and routing structure.

Release notes for MiniMax-M3-uncensored · 428B MoE: the problem, the decision, the artifacts on Hugging Face, and what the evaluation showed. Figures come from the committed snapshot of 2026-09-01.

Problem

Study refusal behavior in a 428B multimodal MoE while preserving its reasoning and routing structure.

Decision

Modified attention o_proj and every residual-writing down_proj across all 60 layers; released the result as 796 GB BF16 in 59 shards.

Artifacts

Artifact	Format	Released	Downloads · 30 days
MiniMax-M3-uncensored (opens in a new tab)	BF16	2026-07-09	637

Result

0/16 hard refusals on the published sample; multimodal behavior, reasoning, and MoE routing were smoke-tested, not fully benchmarked.

Serving check

A minimal single-node vLLM start for the serving build; adjust context length and parallelism to the hardware.

# vLLM, single node; tune --max-model-len and --tensor-parallel-size to the GPU
vllm serve ressl/MiniMax-M3-uncensored \
  --dtype auto \
  --max-model-len 32768

Evidence

Model card: MiniMax-M3-uncensored (opens in a new tab)
Case on ressl.ch: https://ressl.ch/#model-minimax-m3
Hugging Face snapshot captured 2026-09-01