Gryphe/Pantheon-Reasoning-27B

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 29, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Gryphe/Pantheon-Reasoning-27B is a 27 billion parameter Qwen 3.6-based language model developed by Gryphe, featuring a 32768 token context length. This model is an experimental fine-tune focused on enhancing roleplay capabilities through integrated reasoning traces. It is specifically designed to improve character work, narrative planning, and nuanced responses by simulating a writer's thought process before generating text. Pantheon-Reasoning-27B aims to provide a more grounded and prose-forward writing style for interactive fiction and general roleplay scenarios.

Loading preview...

Pantheon-Reasoning-27B: Enhanced Roleplay with Integrated Reasoning

Gryphe/Pantheon-Reasoning-27B is a 27 billion parameter model built on the Qwen 3.6 architecture, specifically fine-tuned to bring advanced reasoning capabilities to roleplay and interactive fiction. This model integrates "thinking traces" into its generation process, simulating a writer's planning phase to improve character consistency, narrative flow, and response quality.

Key Capabilities & Features

  • Reasoning-Enhanced Roleplay: Incorporates explicit reasoning steps for character work, narrative planning, and tone consideration, aiming for more coherent and immersive roleplay experiences.
  • Diverse Training Data: Trained on a wide array of datasets including core Pantheon roleplay data, Opus-4.6-Reasoning-24k (general instruction-following, STEM, coding), WorldSim (long-form narrative roleplay), text adventure content, and general roleplay transcripts.
  • Back-Generated Reasoning: For several datasets, reasoning traces were generated using DeepSeek 3.2, prompting the model to plan its response as a writer would, rather than merely annotating existing text. These traces were validated to ensure genuine forward planning.
  • Uncensored Base Model: Utilizes llmfan46/Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-Preserved as its base, contributing to reduced refusal rates and strong writing capabilities.
  • Preserved Thinking: Trained with preserve_thinking: true, ensuring thinking tags remain active across all assistant turns in multi-turn conversations.

Ideal Use Cases

  • Advanced Roleplay Scenarios: Particularly suited for complex character interactions and narrative-driven roleplay where nuanced responses and consistent character portrayal are crucial.
  • Interactive Fiction & Text Adventures: Excels in generating grounded, prose-forward content for high-stakes interactive storytelling.
  • Experimental AI Research: Serves as a research release to explore the impact of explicit reasoning on roleplay quality and latency, inviting user feedback on its performance.