Experiments
support-agentRun a dataset through two versions and compare quality, cost and speed.
Recent experiments
| Experiment | Compared | Quality | Cost | p95 |
|---|---|---|---|---|
| Refund prompt v19 | v18 → v19 | +4 points | −9% | −0.6 s |
| Cheaper answer model | gpt-4.1 → gpt-4.1-mini | −2 points | −71% | −1.9 s |
| Shorter system prompt | v17 → v18 | +1 point | −12% | −0.3 s |
Sample content. Create a free project to build your own. →