Scale-or-Reason/Qwen2.5-3B-reasoning

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 25, 2025Architecture:Transformer Featherless Exclusive Cold

Scale-or-Reason/Qwen2.5-3B-reasoning is a 3.1 billion parameter Qwen2.5-based student model, distilled from Qwen3-235B-A22B. It is specifically fine-tuned on a large dataset of reasoning answer pairs, making it optimized for tasks requiring complex reasoning. This model is part of a research collection exploring the impact of reasoning in model performance, particularly in general and mathematical contexts. It is designed for applications where explicit reasoning steps are crucial for accurate outputs.

Loading preview...

Model Overview

Scale-or-Reason/Qwen2.5-3B-reasoning is a 3.1 billion parameter student model derived from the Qwen2.5 architecture. It was developed as part of the research paper "When Does Reasoning Matter?" and is specifically designed to excel in tasks requiring explicit reasoning capabilities.

Key Capabilities

  • Reasoning Optimization: This model is a "reasoning model" variant, meaning it has been distilled from Instruction-Fine-Tuned (IFT) or Reasoning answers generated by the larger Qwen3-235B-A22B model. This specialized training focuses on improving its ability to generate logical and step-by-step reasoning.
  • Specialized Datasets: Training utilized extensive datasets of IFT and Reasoning answer pairs, including both a general-reasoning-ift-pairs and a math-reasoning-ift-pairs dataset, highlighting its dual focus on general and mathematical reasoning.
  • Qwen2.5 Base: Built upon the Qwen2.5 family, it benefits from the foundational capabilities of this architecture while being tailored for reasoning-intensive applications.

Good For

  • Research into Reasoning: Ideal for researchers studying the impact and effectiveness of reasoning in language models, as it's part of a controlled study.
  • Applications Requiring Explicit Reasoning: Suitable for use cases where the model needs to demonstrate its thought process or provide step-by-step solutions, rather than just direct answers.
  • Educational Tools: Potentially useful in educational technology for generating explanations or problem-solving steps in general subjects or mathematics.

Finetunes
1 models