Jevgrep Reports 25.8% Lower Total Cost in Updated SWE-bench Coding-Agent Test

Jevgrep, a new open-source retrieval CLI for coding agents, now reports an 8-of-10 SWE-bench result with lower measured task cost by using TypeSafe AI's Jev to decide which repository files and declarations an agent should read first. A fresh total-cost rerun puts the saving at 25.8% when Jev is included, while the project README frames the coding-agent-only result as about 30% lower cost.