Skip to content

EthenEthenEthen

LoginPlatformTry Ethen

Open Source Model Profile · DavidAU

Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking

Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking is a 21.27B-parameter Qwen image-text-to-text fine-tune from DavidAU. According to the model card, it derives from a 27B Qwen 3.5 base with Deckard and Claude-distill training.

Publisher
DavidAU
Task
image-text-to-text
Model type
qwen3_5
License
apache-2.0
Library
transformers
Publication status
Accepted · not indexed

Model overview

Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking is published by DavidAU as a qwen3_5 image-text-to-text model. The captured configuration identifies Qwen3_5ForConditionalGeneration and Safetensors metadata reports 21,268,429,424 parameters. According to the model card, it is a dense 21B variant contracted from a 27B Qwen 3.5 base and trained with Deckard datasets and Claude distill data.

Recorded capabilities

Contracted 21B lineage

According to the model card, the 21B dense model was contracted from a 27B Qwen 3.5 base, then trained with Deckard datasets and a Claude 4.6 Opus distill dataset.

Publisher-reported efficiency

According to the model card, the model has 48 layers and 639 tensors, described as 33 percent smaller and faster than the 27B base with smaller quantized memory use.

Variable-length reasoning

According to the model card, the model uses variable-length reasoning, shorter for less complex prompts and longer for more complex prompts.

Documented sampling guidance

According to the model card, the publisher suggests temperature 0.7 with repetition penalty off for general use, Q6 minimum quants for tool calls, and a system prompt to assist lower quants.

Use cases in the source record

  • Instruction-following and creative-writing experiments consistent with the captured image-text-to-text tag and publisher-described reasoning and storytelling tags.
  • Sampling-sensitive deployment tests using the publisher-documented temperature, repetition-penalty, system-prompt, and quantization guidance.

Limitations and unknowns

  • According to the model card, the publisher describes the model as uncensored and unfiltered, including NSFW responses if requested, so deployment review is needed.
  • No evaluation results were extracted from this record.
  • No context-window, pricing, VRAM-requirement, or latency values were extracted.
  • Provider state is historical snapshot data, not independently refreshed current availability.

Source and provenance

Source: DavidAU/Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking

Captured: Unknown. Processed: 2026-09-07T19:35:38.936111+00:00.

WARNING: This model has character and intelligence. It will take no prisoners. It will give no quarter. Uncensored, Unfiltered and boldly confident. Not even remotely "SFW", if you ask it for NSFW content. And it is wickedly smart too. Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking 21 billion parameters (dense, not moe) CONTRACTED/SHRUNK from 27B Qwen 3.5, then trained on Claude 4.6 Opus High Reasoning dataset via Unsloth on local hardware... but there is much more to the story - in comes DECKARD. 48 layers, 639 Tensors. (33% LESS than base model of 27B) The model is also 33% faster than 27B in terms of token per se…

F001F002F003F004F005F006F007F009F010F011F012F013F014F015F016F017F019F023