Cost & latency
Token and latency accounting — no dollar figure yet
This gateway's per-token pricing isn't public, so token counts and wall-clock latency are reported directly rather than converted to a monetary estimate. No formal quality-vs-cost frontier analysis (bootstrapped, with uncertainty) has been run yet — this is raw accounting, not that analysis.
Baseline 0 · no retrieval
- Input tokens
- 6874
- Output tokens
- 3250
- Total latency
- 74.4s
- Latency / task
- 3.72s
- Failures
- 0/20
Baseline 1 · retrieve + synthesize
- Input tokens
- 45936
- Output tokens
- 5557
- Total latency
- 126.9s
- Latency / task
- 6.35s
- Failures
- 0/20
Agent 1 · planner + retrieve + synthesize
- Input tokens
- 48167
- Output tokens
- 6045
- Total latency
- 160.8s
- Latency / task
- 8.04s
- Failures
- 1/20
Agent1 used ~5% more input tokens, ~9% more output tokens, and ~27% more latency than Baseline1 for the extra planner call — a real, measured overhead, not an estimate. Whether that overhead is worth it depends on the quality comparison on the agents page — where, once graded fairly, it currently isn’t.