Orchard: An open framework for scalable agentic AI
AI Summary: Researchers have introduced Orchard, an open-source framework for scalable and cost-effective agentic AI research. The framework features Orchard Env, a reusable environment service that enables training and evaluating agents across various task domains, including software-engineering, web-navigation, and personal-assistant agents. Orchard has achieved strong results on complex tasks, such as 69.7% on SWE-bench Verified, using relatively small models with approximately 3 billion active parameters. The project also releases training data and evaluation methods to support the broader research community in building and studying open agentic systems.