devwithjevJev Categorizes TODO Titles
Marcell Havlik reports that JEV achieved 97% accuracy categorizing more than 1,000 TODO titles against an Opus 5 reference, compared with 46% for Qwen3-Embedding-4B.
View on X cost10$/million TODO
Marcell Havlik@cviklihamar𝕏
JEV is king at categorization! 🚨 JEV beats the TOP embedder (Qwen3-Embedding-4B) at categorization! 46% vs 97% accuracy! Not a simple win! We benchmarked +1000 TODO titles: - JEV scored 97% accuracy compared to Opus 5 reference. Qwen3 only 46%! - Embedder costed 0.15$/million TODO vs JEV 10$/million TODO Extremly cheap and extreme accuracy while staying superfast! Hopw you guys love it too! Blogpo
The author describes JEV as superfast. The reported cost was 10$/million TODO for JEV, compared with 0.15$/million TODO for the embedder.
Also filed under Triage & routing
- Jev-Assisted Codex Model and Effort Routing
The project provides Jev-assisted model and reasoning-effort routing for local Codex Desktop and CLI.
- Plain-English Gmail Filter Judge
The project uses Jev (TypeSafe) to judge Gmail filters written in plain English.
- Framework-Agnostic Jev Intent and Tool Router
Huzeyfe_dev built a lightweight, framework-agnostic intent and tool router using TypeSafe’s Jev decision model to speed up agent loops.
- MCP Tool Router with Jev and MCTS
Harness-router is a lightweight tool router that uses Jev for fast tool selection and Monte Carlo Tree Search for harder multi-step decisions. It uses MCP as the tool interface and falls back when routing is uncertain.