MoonMath ships uncensored 763B DeepSeek-V4.1 via closed-form weight editing, claims zero capability loss
udmrzn · x · 2026-11-12
MoonMath-ai released DeepSeek-V4.1-Flash-1017-uncensored on Hugging Face (MIT license, 763B MoE), claiming it removed refusal directions via closed-form weight editing rather than lossy fine-tuning, leaving all 384 routing experts untouched. Reported results: near-zero MMLU-Pro drift, GSM8K and GPQA-Diamond scores allegedly exceeding the official base model, and sharply reduced false refusals in coding and reverse-engineering tasks. Ships native FP8 weights with Transformers, vLLM, SGLang and Docker Model Runner support. Claims are unverified by third parties.
More from Models
- Claude Fable 5.2 tops Agent Arena at +30.4%, 1.5x the gain of runner-up Opus 5.5 — daniel_mac8 · 2026-11-12
- Polymarket Puts Only ~40% Odds on Next Gemini Flash (3.9+) Shipping by Year-End — Polymarket · 2026-11-12
- Agent Arena update: Muse Ember (Max) debuts at #25, tied with Muse Wonder — arena · 2026-11-12
- Blood on the Clocktower benchmark: Kimi K3.1 Pro tops open models, just below GPT 5.5 — eliebakouch · 2026-11-12
- Replicating all 9 weight-space programming experiments: big wins, one miss — burny_tech · 2026-11-12
- Open-source Iris-Z unifies image generation and depth in one backbone, beating MoGe-3 — victormustar · 2026-11-12