OpenHands/openhands-lm-32b-v0.1-ep3

TEXT GENERATIONConcurrency Cost:2Model Size:32.8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Mar 25, 2025License:mitArchitecture:Transformer0.0K Open Weights Cold

OpenHands/openhands-lm-32b-v0.1-ep3 is a 32.8 billion parameter open coding model developed by All-Hands-AI, fine-tuned from Qwen Coder 2.5 Instruct 32B. It features a 32768-token context window and is specifically optimized for software engineering tasks, achieving a 37.2% resolve rate on SWE-Bench Verified. This model is designed for local deployment on hardware like a single 3090 GPU, offering strong performance comparable to much larger models.

Loading preview...

OpenHands LM v0.1-ep3: An Open Coding Agent Model

OpenHands LM v0.1-ep3 is a 32.8 billion parameter language model developed by All-Hands-AI, built upon the Qwen Coder 2.5 Instruct 32B foundation. This specific version (ep3) is a research release, trained for three epochs, and is primarily intended for research purposes due to potential issues like looping or not following instructions. For general OpenHands tasks, the all-hands/openhands-lm-32b-v0.1 model is recommended.

Key Capabilities & Features

  • Specialized Fine-tuning: Leverages an RL-based framework (SWE-Gym) using training data generated by OpenHands itself on diverse open-source repositories, focusing on successfully resolved examples.
  • Extended Context Window: Features a 128K token context window, making it suitable for handling large codebases and long-horizon software engineering tasks.
  • Strong Performance: Achieves a 37.2% verified resolve rate on the SWE-Bench Verified benchmark, demonstrating performance comparable to models with 20x more parameters (e.g., Deepseek V3 0324).
  • Local Deployability: Its 32B parameter size allows for local execution on consumer-grade hardware, such as a single 3090 GPU.

Intended Use Cases

  • Research on Coding Agents: Ideal for researchers exploring autonomous agents for software development, particularly in understanding model behavior under extended training.
  • Software Engineering Tasks: Excels at solving GitHub issues and other software development tasks, offering a strong open-source alternative to proprietary models.
  • Local Development: Suitable for developers who require a powerful coding agent that can be run locally without reliance on external API calls.

Finetunes
1 models