Chroma Releases Context-1: A 20B Agentic Search Model for Multi-Hop Retrieval, Context Management, and Scalable Synthetic Task Generation

Context-1 is a 20B-parameter agentic search model that functions as a specialized retrieval subagent.

It is derived from the gpt-oss-20B Mixture-of-Experts base and was fine-tuned by Chroma with CISPO using supervised fine-tuning and reinforcement learning.

The model breaks complex queries into subqueries, executes parallel tool calls (avg. 2.56 calls per turn), and iteratively searches the corpus to locate supporting documents.

Context-1 interacts with tools such as search_corpus (hybrid BM25 + dense search), grep_corpus (regex), and read_document, and includes a Self-Editing Context feature.

The design shifts retrieval logic from the developer to the model to supply relevant documents to a downstream frontier model; pruning accuracy is 0.94.