Augment Code has significantly enhanced its context compression layer by integrating Mercury 2, achieving an 82% reduction in latency and a 90% decrease in costs. The update, implemented on May 13, maintains quality comparable to Opus 4.7. This improvement was realized by fully decoupling and offloading the summarization task to Mercury 2 as a dedicated service, marking a strategic shift in their architecture. The new solution is now live in production, showcasing the potential of Mercury 2 in optimizing performance and cost-efficiency.