Prisma ORM v7.4 ships a new query caching layer that reuses compiled SQL statement structures, reducing per-query compilation cost from 0.1–1ms down to 1–10µs and achieving ~100% cache hit rates in typical applications. The release also adds partial indexes as a preview feature for PostgreSQL, SQLite, SQL Server, and CockroachDB, and fixes BigInt precision issues. The post explains why v7 had performance regressions: moving from Rust/Tokio to a WASM module shifted query compilation onto the main JavaScript event loop, causing contention under high concurrency. Query caching addresses this by parameterizing queries and reusing their static structure. Benchmarks show v7.4 outperforms both v6 and v7.3 in requests per second and latency under load.
