MMOPD/Qwen3-4B-OT3-tau

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

MMOPD/Qwen3-4B-OT3-tau is a 4 billion parameter Qwen3-based language model, fine-tuned for multi-turn, tool-calling agent behavior. It builds upon a 'thinking' model that opens answers with a block, enhancing its reasoning ability with tool-use trajectories from the Tau2 dataset. This model is designed for applications requiring advanced agent capabilities and tool interaction within a 32,768-token context window.

Loading preview...

Overview

MMOPD/Qwen3-4B-OT3-tau is a 4 billion parameter model based on Qwen3-4B-Base, specifically fine-tuned for advanced multi-turn, tool-calling agent behavior. It extends the capabilities of its base model, MMOPD/Qwen3-4B-OT3-2ep, which already incorporated a 'thinking' mechanism (answers begin with a <think> block) for enhanced reasoning. This model was trained on 33,531 Tau2 tool-use trajectories from the inclusionAI/AReaL-tau2-data dataset, focusing on conditioning the model to generate reasoning, visible content, and tool calls based on the conversation history.

Key Capabilities

  • Tool-Calling Agent Behavior: Trained to perform multi-turn tool calls, enabling complex interactive applications.
  • Enhanced Reasoning: Inherits and builds upon a base model designed for explicit 'thinking' processes.
  • Large Context Window: Supports a sequence length of 32,768 tokens, allowing for extensive conversational history and complex interactions.
  • Consistent Training: Developed with identical hyperparameters and data as its smaller counterpart, MMOPD/Qwen3-1.7B-OT3-tau, differing only in model size.

Good For

  • Developing AI agents that require explicit reasoning and tool interaction.
  • Applications needing to process and generate responses within long conversational contexts.
  • Experimenting with models that demonstrate a structured 'thinking' process before generating output.