Thanks for pointing that out. I ran the benchmark with asyncpg and updated the article. It is still slow. I suspect that the JSON serialization is the bottleneck.