Systems Operational
Gateway Benchmarks
AI agents require low latency to maintain reasoning loops. We track our routing overhead globally to ensure we add minimal Time To First Token (TTFT). Numbers are updated continuously using our open-source benchmarking CLI.
12-25ms
Average Gateway Overhead (TTFT)
99.99%
Failover Success Rate
0%
HTTP 429s (With Key Pooling)
Latest Test Results (Q3 2026)
| Model Route | Avg Direct TTFT | Forge Routed TTFT | Added Latency |
|---|---|---|---|
| gpt-4o (us-east) | 315ms | 331ms | +16ms |
| claude-3-5-sonnet (aws-west) | 420ms | 445ms | +25ms |
| llama-3.1-70b (groq) | 115ms | 127ms | +12ms |
Reproducible Methodology
We don't believe in marketing numbers. Our benchmark metrics are generated using an open-source testing harness. The benchmark fires 100 concurrent requests consisting of 50 prompt tokens, measuring the Time to First Token (TTFT) and Time to Final Token.
$ npx @forgeapi/benchmark-cli --target forge --models gpt-4o,claude-3-5-sonnet --concurrency 100