Cheap-first routing: only 4 of 143 questions escalated to a frontier model, cost down 80%
xrplincoln · reddit · 2026-11-16
Instead of sending every request to a frontier model, the author routed per-question: a cheap model (DeepSeek V4.1 Flash, $0.07/M output tokens) answers first, a small checker model scores the answer 0–100 on correctness, instruction-following, and hallucination, and only answers below 75 escalate to a frontier model (Claude Opus 5.5).
Results across 143 questions (104 everyday, 39 stress tests), with costs from actual API token counts:
- Total cost: $0.27 vs $1.33 frontier-only (80% cheaper)
- Average quality: 99.7 vs 98.0
- 139 of 143 questions answered by the cheap model; savings of 98% on everyday questions, 66% on stress tests
The checker escalated 4 questions — three real catches (a wrong count, blank answers on 40-digit multiplication and a hard Sudoku) and one false alarm. In a separate test it caught all 8 deliberately wrong answers.
The weak spot is the checker itself, the only barrier between a bad cheap answer and the user. The project exposes an OpenAI-compatible endpoint, so existing apps only need a base-URL change. Code: github.com/robd-hub/cheap-first.
More from coding & agent
- A satirical year in AI coding: nobody writes code anymore, and we're still figuring out what that means — mattpocockuk · 2026-11-16
- Executor overhauls agent approvals with 4 modes, adds Gmail-style undo for any agent action — threepointone · 2026-11-16
- NVIDIA SkillSpector on 4 popular Agent Skills repos: read the findings, not the score — viest_me · 2026-11-16
- I Shipped an App, a Site, and a CRM With Codex and Claude Code — and Learned Nothing — ThatGuyJord2020 · 2026-11-16
- Over Half of LLM Requests Go to Models Younger Than 60 Days, Vercel Gateway Data Shows — aruthesa · 2026-11-16
- Open-source Craft Anything hits 34K GitHub stars in 16 days by letting Claude Code clone any app — thisguyknowsai · 2026-11-16