Jev agent queries Opus at 4x lower cost
The gist
Two blog tests found pairing TypeSafe's Jev with Claude Opus neared Opus-level accuracy at over 4x lower cost.
The trade-off is a few points of accuracy, and one cascade result is a flagged best case.
How it unfolded
- Nov 3, 2026Blue Orange Digital reports a public OpenRouter test on 3,080 banking support messages (Banking77, 77 intents): Jev sorted 81% correctly versus 84% for Claude Opus, with a median response time of 175 milliseconds versus 2.3 seconds.
- Nov 3, 2026In the same test, Jev cost $0.11 per thousand messages versus $2.42 for Opus with prompt caching and about $19 without it; at a 0.90 threshold a cascade let Jev handle 76% of messages at $0.69 per thousand with 84.0% overall accuracy.
- Nov 4, 2026Hightouch reports testing Jev and GPT-6 Luna on grading identity resolution: both hit 88.8–97.3% accuracy out of the box, and a hybrid Jev + Opus model reached frontier accuracy at more than 4x lower cost.
Key terms
- identity resolution
- The task of finding records across different data sources that refer to the same entity.
Source: Wikipedia