Back to Research
A decision model against a chat model for agent routing
Can a model that writes no text be the router?
Findings
Open a finding to see how it was measured.
- The Jev decision model scored 52 of 52 (lower bound 0.931, p95 389 ms). gpt-4o scored 48 of 52 (p95 2,222 ms), and 1 injection got through.
- In production Jev cost about 18 times less per decision, measured on only 12 decisions.
- Splitting one question into atomic questions raised accuracy from 62.6% to 95%.