Compute Benchmarks
Benchmarks
Reproducible performance measurements across providers. Open methodology, raw data, no marketing spin.
Performance benchmarks for sandbox environments — cold start latency, execution throughput, and reliability across providers.
26 Providers
1 Test Type
Runs Weekly
Object storage benchmarks across providers — upload/download latency, throughput, and reliability.
8 Providers
3 Metrics
Runs Weekly
Headless browser session benchmarks — create, connect, navigate, and release latency across providers.
7 Providers
5 Metrics
Runs Weekly
Cold and warm connection-phase latency, streaming throughput, and reliability across AI gateways, measured against a direct-to-provider baseline for four target models.
3 Gateways
4 Model Families
Runs Weekly
Need independent, third-party verification for your own claims? We'll design and run a benchmark scoped to your infrastructure and workload.
Last benchmark run: August 14, 2026 · View raw data on GitHub ↗