osmapi/osmQwopus3.6-27B-Fable-Agentic
The osmapi/osmQwopus3.6-27B-Fable-Agentic is a 27 billion parameter agentic model developed by osmapi, fine-tuned from Jackrong/Qwopus3.6-27B-v2. It excels in step-by-step reasoning, multi-turn tool-calling, and provides top-tier general knowledge, achieving 88.29% on MMLU-Pro and 97.33% on GSM8K. This model is designed for complex tasks requiring advanced reasoning and tool interaction.
Loading preview...
Overview
osmQwopus3.6-27B-Fable-Agentic is a 27 billion parameter agentic model, fine-tuned from Jackrong/Qwopus3.6-27B-v2. It is designed for advanced reasoning and multi-turn tool-calling, while maintaining top-tier general knowledge. The model demonstrates improved knowledge over its base model, particularly in MMLU-Pro, and preserves strong performance in mathematical and complex scientific reasoning tasks.
Key Capabilities
- Agentic, multi-turn tool-calling: Enables interaction with external tools over extended sessions.
- Step-by-step reasoning: Utilizes chain-of-thought for complex problem-solving.
- Top-tier general knowledge: Achieves 88.29% on MMLU-Pro, showing a +1.43% improvement over its base model.
- Strong mathematical ability: Scores 97.33% on GSM8K.
- Multimodal architecture: Supports both vision and text inputs.
Benchmarks & Performance
This model surpasses its base on MMLU-Pro (88.29% vs 86.86%) and maintains comparable performance on GSM8K (97.33% vs 98.00%) and GPQA-Diamond (57.07% vs 57.58%). Notably, it shows significant improvement in business (+10%) and health (+4%) categories within MMLU-Pro. The model is optimized to provide sufficient context for detailed reasoning, with an 8192-token budget used during evaluation.
Usage Tip
When using this model, it is recommended to provide a generous max_tokens setting to allow the model to complete its full chain of thought during reasoning processes.