philbert440/ThinkingCap-Qwen3.6-27B-Uncensored-Medium

VISIONPricing:Input $1.06 / Cached $0.15 / Output $2.6Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 1, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

philbert440/ThinkingCap-Qwen3.6-27B-Uncensored-Medium is a 27 billion parameter Qwen3.6-based vision-language model, part of the ThinkingCap collection. This 'Medium' tier variant offers a balanced openness profile, providing sharp performance (gsm8k 0.94–0.95, factual 1.0) with selective uncensoring for cyber/NSFW content. It is provided as a BF16 full-precision checkpoint, suitable for further quantization or use on GPUs with ample memory, and is optimized for scenarios requiring a nuanced approach to content generation.

Loading preview...

ThinkingCap Qwen3.6 27B Uncensored Medium

This model is a 27 billion parameter variant from the Qwen3.6-27B Uncensored — ThinkingCap collection by philbert440. It is based on the bottlecapai/ThinkingCap-Qwen3.6-27B model and is designed for users seeking a balance between performance and content openness.

Key Characteristics

  • 'Medium' Tier Abliteration: Represents the sharpest and best-calibrated tier within the ThinkingCap series, achieving strong performance metrics (gsm8k 0.94–0.95, factual 1.0). It features 'gentle abliteration,' meaning it has selective openness.
  • BF16 Full-Precision Source: Provided as a full-precision BF16 checkpoint, making it suitable for users who wish to perform their own quantization or have GPU resources to run it in higher precision.
  • Selective Uncensoring: The model's 'uncensored' nature specifically permits prompts related to hacking, malware, lock-picking, NSFW content, and disinformation. However, it maintains refusals for topics such as weapons, drugs, political persuasion, surveillance, and extremism, aiming for a responsible yet open profile.
  • Vision-Language Capabilities: As a Qwen3.6-based model, it inherently supports vision-language tasks, though specific details on vision performance are not provided in this README.

Use Cases

This model is well-suited for applications requiring a high-performing language model with a more permissive stance on certain sensitive topics (cybersecurity, NSFW) while still adhering to ethical boundaries regarding harmful content like weapons or extremism. Its BF16 format offers flexibility for deployment and further optimization.

Finetunes
2 models