open-thoughts/OpenThinkerAgent-32B

TEXT GENERATIONConcurrency Cost:2Model Size:32BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 8, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

OpenThinkerAgent-32B is a 32 billion parameter language model developed by OpenThoughts, post-trained from Qwen3-32B. It is specifically fine-tuned for agentic tasks using the 100,000-example OpenThoughts-Agent-SFT-100K dataset. This model excels in agentic benchmarks, demonstrating strong performance across a suite of seven agentic tasks, making it suitable for complex problem-solving and automated agent applications.

Loading preview...

OpenThinkerAgent-32B: A Specialized Agentic Model

OpenThinkerAgent-32B is a 32 billion parameter model developed by OpenThoughts, specifically designed for agentic applications. It is post-trained from the Qwen3-32B base model using a comprehensive 100,000-example dataset, OpenThoughts-Agent-SFT-100K. This dataset comprises (task, agent-trajectory) pairs derived from top task sources like SWE-Smith and StackExchange, with trajectories generated by GLM-4.7-AWQ and filtered for multi-turn interactions.

Key Capabilities

  • Agentic Task Performance: Achieves an average score of 44.8 across seven agentic benchmarks, including SWE-Bench-Verified, Terminal-Bench 2.0, and MedAgentBench.
  • Strongest Open-Data 32B Model: Positioned as the leading open-data model in its size class for agentic benchmarks.
  • Robust Training: Utilizes full-parameter SFT with a cutoff length of 32768 and bf16 precision, ensuring high-quality instruction following.

Good For

  • Automated Problem Solving: Ideal for tasks requiring autonomous agents, such as code generation, debugging, and complex multi-step reasoning.
  • Research and Development: Provides a strong foundation for further research into agentic AI and dataset curation.
  • Benchmarking Agentic Systems: Serves as a high-performing baseline for evaluating new agentic models and methodologies.