What it does
On 751 replayed routing decisions, Jev picked the right expert more often than the prompt-based router and answered 8.1× faster at the median. Reranking was the reality check: 40.2% Recall@6 against 44.3% for production, at roughly $0.00044 a query, with tie-breaking doing part of the work. Citation judgments repeated more consistently than GPT-5.6 Luna's. Their advice — benchmark it wherever the answers can be listed in advance — is the most reusable sentence in the piece.
Maker-reported (not independently measured by JevMade): Author-reported (Glean): 8.1× median per-entry speedup over the prompt-based router, with higher accuracy on 751 golden entries · Author-reported (Glean): Jev Choice 40.2% vs production 44.3% Recall@6 on 4,855 queries, about $0.00044 per query, 0.195s p50 · Author-reported (Glean): one changed citation-coverage verdict across three identical runs, versus changes under both GPT-5.6 Luna configurations
- Primitives
- Not stated
- Added
- Project created
Source checked 2026-09-29 — opened the primary source directly.