#rlvr

Explore why the Chroma Context-1 agentic search model excels at multi-domain retrieval. How can a compact model beat trillion-parameter giants? Context-1 utilizes Reinforcement Learning from Verifiable Rewards (RLVR), teaching the system to aggressively prune its workspace. By being trained to constantly edit out unnecessary contexts (94.1% accuracy rate), it maintains high fidelity without context rot. Learn about its incredible speed, as it achieves a remarkable 2.56 tool calls on average. With a 0.98 F1 score in out-of-domain email search, Context-1 (4x) completely outperformed GPT-5.4 and Opus 4.6 on the BrowseComp+ benchmark.