Bifrost history
Aug 2026 → Nov 2026
openai → openai cell. The engine changed between these runs (f3adbb1 → d240bad), so differences can include instrument changes, not only gateway changes. Changes within ±5% show as roughly unchanged (—); this is a display tolerance, not a statistical margin.
| Reading | Aug 2026 | Nov 2026 | Change | Rank Aug 2026 → Nov 2026 |
|---|---|---|---|---|
| Added latency p50 (µs) | 526 µs | 658 µs | ↓ +25.1% | 8 of 14 → 9 of 15 |
| Added latency p99 (µs) | 1,029 µs | 1,057 µs | — +2.7% | 8 of 14 → 10 of 15 |
| Throughput, p99 ≤ 5 ms (req/s) | 4,570 req/s | 4,346 req/s | — -4.9% | 1 of 1 → 1 of 1 |
| Throughput, p99 ≤ 10 ms (req/s) | 4,791 req/s | 4,555 req/s | — -4.9% | 1 of 1 → 1 of 1 |
| Throughput, p99 unbounded (req/s) | 5,535 req/s | 4,624 req/s | ↓ -16.5% | 1 of 2 → 2 of 3 |
| CPU per request (µs) | 806.9 | 865.1 | ↓ +7.2% | 9 of 14 → 11 of 14 |
| Idle RSS (MiB) | 216.9 MiB | 160.3 MiB | ↑ -26.1% | — → — |
| Peak RSS (MiB) | 906.4 MiB | 1,257 MiB | ↓ +38.7% | — → — |
Runs
2 published runs · Added latency p99 (µs), openai → openai · values describe the recorded local-mock configuration; a missing reading is never zero. Hover a trend for its comparability details.
| Measured | Build | Engine | Result | Trend | Field placement | Evidence |
|---|---|---|---|---|---|---|
| maximhq/bifrost:v2.5.0 | d240bad | 1,057 µs | Engine changed | 10 of 15 Current field | Run evidence | |
| maximhq/bifrost:v1.6.11 | f3adbb1 | 1,029 µs | First run | 8 of 14 Aug 2026 field | Run evidence |