mhseo2/Qwen3-4B-OT3-2ep-opd

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 16, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

mhseo2/Qwen3-4B-OT3-2ep-opd is a 4 billion parameter Qwen3-Base model fine-tuned on the OpenThoughts3-1.2M dataset, which comprises 1.2 million long-chain-of-thought examples covering math, code, and science reasoning. This model is specifically designed as a "thinking" model, where every answer begins with a dedicated thinking block, making it highly effective for complex reasoning tasks. It serves as a foundational general reasoning student within the MMOPD project, intended for further distillation into domain-specific teachers.

Loading preview...

Overview

mhseo2/Qwen3-4B-OT3-2ep-opd is a 4 billion parameter model based on Qwen3-4B-Base, fine-tuned for 2 epochs on the extensive OpenThoughts3-1.2M dataset. This dataset consists of 1.2 million long-chain-of-thought examples, focusing on math, code, and science reasoning traces. The model is engineered to be a "thinking" model, meaning every generated response initiates with a <think> block, explicitly outlining its reasoning process. This design makes it particularly adept at tackling complex reasoning problems.

Key Capabilities

  • Enhanced Reasoning: Specialized in long-chain-of-thought reasoning across mathematics, coding, and scientific domains.
  • Explicit Thinking Process: Generates a <think> block before the final answer, providing transparency into its problem-solving steps.
  • Foundation for Distillation: Serves as a general reasoning student within the MMOPD (multi-teacher on-policy distillation) project, acting as a starting point for creating domain-specific teacher models.
  • Competitive Benchmarks: Achieves notable scores on reasoning benchmarks such as AIME24 (66.3%), LiveCodeBench v6 (51.7%), and IFEval (51.0%), demonstrating strong performance in its specialized areas.

Good for

  • Complex Problem Solving: Ideal for applications requiring detailed, step-by-step reasoning in technical and academic fields.
  • Educational Tools: Can be utilized in systems that benefit from explicit thought processes, such as tutoring or explanation generation.
  • Research in Model Distillation: A valuable base model for researchers exploring multi-teacher on-policy distillation techniques and creating specialized AI agents.

Finetunes
1 models