YOU ARE VIEWING ONE ITEM FROM THE AICRIER FEED

Darkc0de's Heretic-tuned 119B Mistral Small 4 drops quietly

AICrier tracks AI developer news across Product Hunt, GitHub, Hacker News, YouTube, X, arXiv, and more. This page keeps the article you opened front and center while giving you a path into the live feed.

// WHAT AICRIER DOES

7+

TRACKED FEEDS

24/7

SCRAPED FEED

Short summaries, external links, screenshots, relevance scoring, tags, and featured picks for AI builders.

Darkc0de's Heretic-tuned 119B Mistral Small 4 drops quietly
OPEN LINK ↗
// 64d agoMODEL RELEASE

Darkc0de's Heretic-tuned 119B Mistral Small 4 drops quietly

Darkc0de's Heretic build is a decensored BF16 fork of axolotl-ai-co/Mistral-Small-4-119B-2603-BF16, and the Reddit post says the real blocker for local users is still a GGUF build. The model card claims far fewer refusals than the original, but the thread is still tiny and early.

// ANALYSIS

Interesting niche release: this is mostly about behavior tuning, not a new architecture, and it only matters if the community turns it into a usable local package. Right now it's a power-user curiosity, not a default recommendation.

  • Built with Heretic v1.2.0 as a decensored fork of the official Mistral Small 4 119B base.
  • The model card says refusals fell from 98/100 on the original to 27/100 here, so the selling point is behavior, not capability jumps.
  • The 119B BF16 footprint is heavy for local runs, which makes GGUF support the practical adoption gate.
  • The visible Reddit comments skew skeptical, with criticism of coding quality and of the whole abliterated-model idea.
  • Only 520 downloads last month and no inference-provider deployment suggest this is still very early.
// TAGS
llmopen-weightsself-hostedmistral-small-4-119b-2603-heretic

DISCOVERED

64d ago

2026-03-24

PUBLISHED

64d ago

2026-03-24

RELEVANCE

8/ 10

AUTHOR

Quiet-Owl9220