Microsoft Research has announced Orchard, an open-source framework intended for the research community to train and evaluate AI agents across a range of task types. According to the announcement, the framework reduces complexity by allowing researchers to reuse the same underlying infrastructure. It is also described as supporting strong performance from smaller models.
Why it matters
Shared, reusable infrastructure for training and evaluating agents can lower the barrier to research and make experiments more consistent across task types. The stated emphasis on enabling smaller models to perform well points to a focus on efficiency alongside capability.
Who should care
Researchers working on agentic AI and model evaluation are the primary audience, as the framework is aimed at the research community and released as open source.