Skip to main content

Benchmarks

Request runtime​

benchmarks/runtime_benchmark.py compares a shared, concurrent-safe service using one extraction context per build with constructing a complete service for every build. It uses an offline two-item HTML fixture, eight worker threads, 50 builds per sample, and performs no network or durable-cache I/O.

python benchmarks/runtime_benchmark.py

The result recorded on 2026-08-04 (Python 3.13.7, Apple Silicon) was:

  • Shared request contexts: 543.2 builds/second (median of three samples).
  • Isolated service per build: 39.1 builds/second.
  • Shared/isolated throughput ratio: 13.89x.

Numbers are directional and should be compared on the same host; the regression signal is the shared/isolated ratio rather than absolute builds per second.

Browser rendering​

python benchmarks/browser_benchmark.py

See browser rendering for the measured baseline and concurrency guidance. See performance for day-to-day extraction advice (specs, caches, pattern reuse).