Drex Router: a small model picks the right LLM per step, 28x cheaper at near-equal quality
aftahi_ai · x · 2026-11-20
Drex Router is a small model that routes each agent step to the right LLM/SLM and reasoning effort instead of defaulting to the most expensive one. Benchmarks: 8.80/10 judge quality at $0.28 per 1k prompts vs Jev Router's 8.84 at $7.80 (28x cheaper), and 87.6% MMLU accuracy at $0.10 per 1k questions vs Jev Router's 86.4% at $1.09, while a 12B base model scores 74.1%. Try it online; commenting "DREX" gets $5 in API credits.
Related event: Nace.AI's Drex Router Picks the Best LLM for Every Agent Step(2 posts)→
More from coding & agent
- BMW engineers assign trust per step in agent workflows, with code-enforced permission checks — shoumikgoswami · 2026-11-20
- Claude Code's creator posts free 28-minute video on prompting, CLAUDE.md and parallel sessions — aftahi_ai · 2026-11-20
- Google Antigravity users complain about opaque 10-hour runs and laggy side chat — jasondeanlee · 2026-11-20
- Is AI overengineering incompetence or just different priors? Devs debate model coding habits — austinc3301 · 2026-11-20
- Dev rewrites 100M+ LOC in strict memory-safe Rust, betting everything will be AI-rewritten — doodlestein · 2026-11-20
- Running parallel coding agents on one repo: a checkpoint handoff system and where it breaks — linhtoji · 2026-11-20