RESEARCH · BENCHMARKS
Numbers from live deployments.
Reproducible benchmarks from production. Methodology on request.
BENCHMARKS · Q1 2026
Numbers from live deployments.
RAG HALLUCINATION
− 68 %
Baseline → live
AGENT RESPONSE p95
(PRODUCTION LOAD)
< 90 ms
Down from ~800 ms
QML TRAIN TIME
(vs CLASSICAL)
97 ×
Weeks → hours
Full methodology available on request.