OrcaRouter Open-Sources Abliterated DeepSeek V4.1 Flash GGUF Weights

X postOrcaRouterAnnouncement

OrcaRouter has open-sourced abliterated GGUF weights for DeepSeek V4.1 Flash 1017, claiming a significant drop in refusals while maintaining core reasoning benchmarks.

Key points

  1. Release of OrcaCyber DeepSeek V4.1 Flash 1017 as an uncensored/abliterated GGUF model on Hugging Face.
  2. OrcaRouter reports 91.5% fewer refusals compared to the standard model.
  3. MMLU score reported at 92%, compared to 91% for the original weights.
  4. Configured to run locally on systems with 128 GB or 256 GB RAM without cloud reliance.
  5. OrcaRouter noted in follow-ups that its Cyber tier hosts full-precision versions.

Why it matters

Abliteration removes refusal vectors from open-weight models to prevent overly cautious rejections in specialized or automated pipelines. Providing these as GGUF quantizations allows technical teams with high-spec local hardware to run unconstrained inference independently of third-party cloud APIs.

Analysis

By stripping refusal directions while retaining benchmark performance on MMLU, OrcaRouter targets technical environments requiring unrestricted model outputs. However, with hardware requirements around 128 GB to 256 GB of memory, practical local execution remains limited to enterprise-grade workstations or top-tier unified memory setups until smaller quantizations are released.

Original post

OrcaRouter 🐳

@OrcaRouter · 24 October 2026

We're open-sourcing our OrcaCyber DeepSeek V4.1 Flash 1017 (Abliterated GGUF) weights — one of the strongest uncensored/abliterated DeepSeek V4.1 weights we've tested, with virtually no capability loss. - 91.5% fewer refusals - 92% MMLU vs. 91% original (slight improvement) - t.co/Eeljn12BoA

3.5k likes77 replies3.4k bookmarks250.7k views

View on X
More posts in the thread (1)
  1. @xie_uh Cyber tier hosts full precision models
Selected replies (5)
  • @MaxRovensky @OrcaRouter > Runs locally on 128GB my MacBook about to go brrrrrrr
  • @NaraharikripaM @OrcaRouter @MiaAI_lab @ashxhart can we have this model for 2dgx on tensorfold ?
  • @xie_uh @OrcaRouter What's the gap between cyber tier and hugging face ?
  • @MgkMshrmBrkfst @OrcaRouter would this work for strata?
  • @mimies_tech @OrcaRouter 235.8 GB is not local for most people. Even the 114 GB build needs a big Mac Studio. Real local use needs a smaller quant that keeps KLD near 0.05.

Mentioned links

Links and tools from this entry

Summary generated by AI (Gemini 3.8 Flash, 27 October 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Link:OrcaRouter

    OrcaRouter Releases Abliterated FP8 Build of Qwen3.8-27B

    OrcaRouter has released an abliterated, block-FP8 quantized version of the multimodal model Qwen3.8-27B on Hugging Face. By orthogonalising the internal refusal direction, standard safety guardrails were removed specifically for red-teaming, interpretability, and AI safety research.

    AI & LLMs· Tool

  • X post:Mia

    DeepSeek Releases Official DeepSeek-V4-Flash-0731 Weights on Hugging Face

    DeepSeek has officially published the weights for DeepSeek-V4-Flash-0731 under an MIT licence on Hugging Face. The release supersedes the preview version, features enhanced agentic capabilities, and incorporates an attached speculative decoding module.

    336bookmarks153.4kviews

    AI & LLMs· Announcement

  • Link:Continuum AI

    OrcaRouter on Hugging Face: Security-First AI Gateway and Open-Source Models

    Continuum AI hosts its open-source tools, security research datasets, and quantized model builds on Hugging Face under the OrcaRouter organization. The project pairs a privacy-focused model gateway with tooling for agent observability, post-quantum encryption, and adversarial safety evaluation.

    AI & LLMs· Resource

  • X post:Samuel Cardillo

    OrcaRouter Releases Uncensored Weights for Qwen3.8 27B FP8

    Samuel Cardillo highlights OrcaRouter's release of uncensored weights for Qwen3.8 27B FP8, suggesting Anthropic will be furious. According to OrcaRouter, the weights are designed specifically for AI red teaming and security research.

    604bookmarks227.9kviews

    AI & LLMs· Opinion

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.