Open-source Jev router cuts LLM token costs ~9x by routing Claude Code and Codex traffic
airesearch12 · x · 2026-09-26
A developer released whichmodel.app, a free MIT-licensed LLM auto-router built on Jev-class models, claiming 9x token cost savings with no noticeable quality drop.
Key points:
- One-line install (curl ... install.sh | sh), or point your agent at the repo; sits in front of Claude Code, Codex, opencode, or Cursor
- Uses quota included in your plans when it fits, sends easy turns to cheaper models on your own keys, and can route to free local models
- Routing decision takes 0.5s: every candidate is priced for the actual request (including redo risk), cheapest expected total wins, and cheap answers are checked by Jev before delivery
- Leverages benchmark and per-task cost data from benchmarkheaven
A subtle detail most routers miss: warm cache matters — a provider that already read your conversation bills the next turn at 1/10th input price, so the real question each turn is whether a better model is worth abandoning a warm cache.
More from coding & agent
- dhh: Human-Only Code Review Will Soon Be Seen as Deeply Irresponsible — deliprao · 2026-09-26
- Grok bot packs 750GB of open-source AI models into one downloadable archive — RachelVT42 · 2026-09-26
- Codex is down, so this researcher built an AI chain to watch it come back — Yuchenj_UW · 2026-09-26
- Claude Opus 5.5 one-shots a promo video in Cursor: has video animation been solved? — JosephJacks_ · 2026-09-26
- dhh: Opus 5.5 one-shot ported Rust screensaver engine to x86-64 asm, 17x faster — zack_overflow · 2026-09-26
- Anthropic details path to recursive self-improvement as engineers ship 8x more code — dlwh · 2026-09-26