Prisma 7 is still the version to use in production today. Prisma Next is in Early Access; once it's Generally Available you'll be able to run it in production too.
Peak Sustained Throughput
Mean of each setup's busiest 10% of the load test (higher is better)
p95 Latency Under Load
Traffic ramps up over the run — Prisma 7 saturates early, Prisma Next holds out far longer (lower is better)
Latency at the Same Load
6,000–7,000 req/s — a level all three setups can serve, so the comparison is like for like (lower is better)
| pg (raw driver) | Prisma Next | Prisma 7 | |
|---|---|---|---|
Average  Mean latency across all requests at this load. | 0.7ms | 1.2ms | 8.1ms |
| p95 95% of requests complete faster than this — the slowest 5% take longer. | 1.9ms | 4.4ms | 41.8ms |
| p99 99% of requests complete faster than this — the slowest 1% take longer. | 4.3ms | 8.9ms | 116.7ms |
The Same Query, Three Ways
The paginated-customers endpoint from the benchmark suite, as each setup implements it
Try Prisma Next
Scaffold a complete template app, or add Prisma Next to an existing one:
bunx create-prisma@nextbunx prisma-next@latest initThe benchmark is a fork of the open-source drizzle-benchmarks suite: the identical Postgres workload runs against each setup on the same machine, ramping traffic until each reaches its limit.
Read the full result analysis on the blog · Run the benchmark yourself