Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2
Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2 is a 26 billion parameter language model based on the Gemma 4 MoE architecture, specifically fine-tuned for enhanced reasoning capabilities within roleplay scenarios. This model integrates back-generated thinking traces into its training, allowing it to plan narrative beats and character responses. It is designed to improve the quality and depth of interactive fiction and roleplay by simulating a writer's planning process.
Loading preview...
Model Overview
Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2 is a 26 billion parameter model built on the Gemma 4 MoE architecture, representing a refresh of the original 26B-A4B finetune. Its core innovation lies in integrating reasoning traces into its training, specifically designed to enhance roleplay quality by simulating a writer's thought process before generating responses. This version, 1.1, features tightened reasoning traces and a refined training recipe, focusing on higher quality data.
Key Capabilities & Features
- Enhanced Reasoning for Roleplay: Trained with back-generated thinking traces that guide the model to plan character responses, weigh tone, and consider narrative direction.
- Gemma 4 MoE Base: Utilizes the Gemma 4 Mixture-of-Experts architecture, making it a more efficient choice for training compared to larger models.
- Specialized Training Data: Incorporates Pantheon roleplay data, general roleplay transcripts, and text adventure content, all augmented with reasoning traces.
- Writer-like Planning: Thinking traces are generated by prompting a model to think "as a writer planning their next response," ensuring genuine forward planning rather than post-hoc analysis.
- Improved Writing Quality: Metrics show a significant reduction in clichés and more condensed reasoning traces compared to base models.
Ideal Use Cases
- Interactive Fiction & Text Adventures: Excels in generating grounded, prose-forward content for high-stakes interactive narratives.
- Advanced Roleplay Scenarios: Designed for complex roleplay where character consistency, nuanced responses, and narrative planning are crucial.
- Research into Reasoning in LLMs: Serves as a research release to explore whether explicit reasoning truly enhances roleplay quality and user experience.