AI BriefingPantheon-Reasoning-27B Released
Key point
Pantheon-Reasoning-27B, a roleplay-specialized model that integrates reasoning capabilities based on Qwen 3.6 27B, has been unveiled.
Details
Pantheon-Reasoning-27B is an uncensored model based on Qwen 3.6 27B, designed to leverage reasoning traces for character tone, narrative planning, and response prediction during roleplay.
The key training data composition is as follows:
- Pantheon data (~28%): The core roleplay corpus with reasoning traces backward-generated and applied
- Opus-4.6-Reasoning-24k (~21%): General instruction-following and STEM/coding data using Claude Opus 4.6's reasoning traces
- WorldSim data (~16%): Roleplay data for long-form storytelling and character immersion
- Text adventure data (~16%): Interactive fiction content
This model uses llmfan46/Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-Preserved as its base model, with a focus on reducing the model's refusal responses and maximizing writing ability.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Escha-W2 (Hugging Face repository)