AMAImedia/Qwen3.8-27B-Abliterated-Uncensored-NOESIS-BF16

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 31, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The AMAImedia/Qwen3.8-27B-Abliterated-Uncensored-NOESIS-BF16 is a 27 billion parameter multimodal large language model, derived from Qwen/Qwen3.8-27B and developed by AMAImedia as part of the NOESIS platform. This BF16 model is specifically modified to significantly reduce refusal behavior across a wide range of prompts while retaining its hybrid text and vision capabilities. It is designed for applications requiring highly compliant and versatile AI responses, supporting a 32768 token context length and a broad array of languages.

Loading preview...

Model Overview: Qwen3.8-27B-Abliterated-Uncensored-NOESIS-BF16

This model, developed by AMAImedia as part of the NOESIS Professional Multilingual Dubbing Automation Platform, is a 27 billion parameter multimodal derivative of Qwen/Qwen3.8-27B. Its primary distinction is a significant reduction in refusal behavior, achieved through modifications to the original model. It maintains the upstream hybrid text-and-vision backbone and supports a 32768 token context length.

Key Capabilities & Differentiators

  • Reduced Refusal Behavior: Achieved 0 refusals across an internal 842-prompt screen and a separate 126-prompt holdout, indicating high compliance. This makes it suitable for applications where avoiding refusals is critical.
  • Multimodal Support: Retains the ability to process both text and vision inputs, with the model.safetensors file providing the validated multimodal path.
  • Multilingual: Supports a wide array of languages, as indicated by the extensive language tags.
  • Coherence: Passed 23 out of 24 coherence checks for tasks like coding, JSON, debugging, explanation, and math.
  • BF16 Precision: Utilizes BF16 precision, with the full model requiring approximately 56 GB of VRAM.

Use Cases & Considerations

This model is particularly well-suited for applications demanding high compliance and a low refusal rate, especially in multilingual and multimodal contexts. Developers can leverage its text and vision capabilities for diverse tasks. While internal evaluations show strong refusal reduction and coherence, users should note that these are automated internal metrics and not public leaderboard results. The "uncensored" aspect refers to refusal reduction and is not a guarantee against all undesirable outputs. The model's reasoning, factuality, coding, or vision benchmark scores are not claimed to be preserved unchanged from the base model.

Finetunes
1 models