TeichAI/Gemma-4-31B-Fable-5-Agent-Distill
TeichAI/Gemma-4-31B-Fable-5-Agent-Distill is a 31 billion parameter instruction-tuned causal language model based on the Gemma 4 architecture. Developed by TeichAI, this model is specifically distilled from Claude-code traces and community chat data, making it highly optimized for advanced code generation, scientific reasoning, and complex research tasks. It excels at instruction-following for tasks requiring high logical coherence, leveraging its robust base model and reasoning distillation.
Loading preview...
Overview
TeichAI/Gemma-4-31B-Fable-5-Agent-Distill is a 31 billion parameter instruction-tuned model built upon the Gemma 4 architecture. It has been specifically trained by TeichAI using a distillation process from Claude-code traces, augmented with community-provided chat data. This training methodology, particularly the high-effort reasoning distillation, aims to enhance its capabilities in complex problem-solving and logical coherence.
Key Capabilities
- Coding: Advanced code generation, debugging, and software architecture planning.
- Science: Deep scientific reasoning, hypothesis evaluation, and analytical problem-solving.
- Deep Research: Navigating complex, multi-step research queries and synthesizing vast amounts of information.
- General Purpose: Highly capable instruction-following for everyday tasks requiring strong logical coherence.
Usage Best Practices
For optimal performance, users are recommended to employ specific sampling parameters: temperature=1.0, top_p=0.95, and top_k=64. The model also supports a "Thinking Mode" which can be enabled or disabled using the <|think|> token in the system prompt, allowing for internal reasoning output. Benchmarks for this specific distilled model are anticipated soon, with reasoning capabilities inherited from the base Gemma 4 architecture left untouched.