Blog

A leaderboard scores someone else's task. Build a 20 to 50 example held-out set from your own inputs, pick a grader that matches the work, and score candidates on cost per correct answer. Runnable harness included.

Stop judging routers by the pricing page. Point a trial key at one and read what it shows you about a real request. Seven criteria, each with a five-minute test, and a scorecard you fill in yourself.

Point your OpenAI transcription calls at FluxRouter, use flux-voice, done. Same key, same base URL, one new model id. Python + curl, request fields, response formats, costs and limits. Runs on any paid plan.

Turn a live web page into LLM-ready markdown in one call. POST /v1/fetch, same Flux key and base URL as your chat calls, one field for the URL, one flag for JS-heavy pages. curl + Python, request and response shape, costs and limits. Runs on any paid plan.

Every LLM call is a bet on the same three corners: latency, cost, quality. You can buy two. The real skill is choosing which one to sacrifice, per workload, not once for your whole app.

What LLM routing actually is, the four jobs a router does for you, when routing is the wrong call, how a router differs from a gateway, and how to trust and evaluate any router by what it shows you about one live request.

Switch from OpenRouter to FluxRouter in three lines: new base URL, new key, new model id. Same OpenAI SDK. Mapping table + cutover test plan.

A 107x price gap between models means most of your API calls pay a premium they never earned. Right-size each prompt and the bill follows the work.

The price-quality frontier moves monthly and your hardcoded default never notices. Decide the model per request instead of baking one into a config file.

The monthly invoice hides everything that matters: which model answered, what each call cost, how much was failed retries. You can't cut what you can't see.