Microsoft Research has released Orchard, an open-source framework designed to enable scalable and cost-effective research into autonomous agents. The system centers on Orchard Env, a Kubernetes-based environment service that allows researchers to train and evaluate agents across diverse domains like software engineering, web navigation, and personal assistance without rebuilding infrastructure.
- Orchard-SWE achieves 69.7% on SWE-bench Verified using ~3 billion active parameters, reaching 73% with value-model reranking.
- Orchard-GUI trains a 4-billion-parameter vision-language model to achieve an average of 68.4% across WebVoyager, Online-Mind2Web, and DeepShop benchmarks.
- The framework supports training agents directly inside real deployment harnesses like Codex and OpenClaw, closing the gap between training and deployment environments.
The release includes training data and evaluation methods to help the broader research community build and study open agentic systems, addressing the bottleneck of proprietary infrastructure in current agentic AI development.