Flux RouterDocumentation

Let Flux pick, or pin a model

The first choice you make: let flux-auto pick a model for each request, or pin one model so the same one answers every time.

You set the model on every request with the model field. With Flux you have a choice: let flux-auto pick a model for you, or pin a specific model so the same one answers every time. This page helps you decide.

What is the difference?

  • flux-auto picks among vetted models for each kind of task, cost-aware. You do not name a model; Flux picks one. Best when you want good results across a mix of requests without managing model selection. See How routing works.
  • A lane alias (flux-fast, flux-standard or flux-reasoning) keeps Flux picking the model, but only from that lane. See Lanes.
  • Pinning means passing a flux-pinned-* alias so one exact backing model serves the request every time. You give up the automatic pick in exchange for a guarantee about which model you get.

All three go through the same gateway on the same key, so your agent's setup does not change.

What each choice costs

  • flux-auto bills at the Standard rate, or the Fast rate when it sends your request to the Fast lane and a Fast model answers. Never more than Standard. Send flux-fast if you want the Fast rate on every request.
  • Lane aliases are billed at their lane's rate.
  • Pinned models are billed at the provider's rate plus the pay-as-you-go markup.

Current rates are on Pricing. How billing works has the detail.

When should I pin a model?

Pin when you need a property that only a specific model can guarantee:

  • Determinism and reproducibility. If you are running evals, regression tests, or anything where the same input must hit the same model, pin it. With flux-auto the pick is made per request; pinning removes that variable.
  • Capability guarantees. If your request needs a feature only some models support, pin a model you know supports it.
  • Output consistency. Different models have different styles and formats. If you have tuned prompts against one model's behaviour, pin it so that behaviour does not change underneath you.
  • Long context. If you need a large context window, pin a long-context model. See Context windows and long inputs.

When should I use flux-auto?

Use flux-auto (or a lane alias) when you want Flux to size each request and you do not need a guarantee about the exact model:

  • Agents doing a mix of simple and hard work.
  • General assistant, chat and content tasks where any capable model is fine.
  • Work where you would rather not track which model is best this month.

How do I pin a model?

Set a flux-pinned-* alias as the model in your agent's config, or as the model in a request. Everything else stays the same.

bash
curl https://api.fluxrouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "flux-pinned-...",
    "messages": [{ "role": "user", "content": "Run this regression prompt." }]
  }'

The authoritative list of flux-pinned-* ids is GET /v1/models, and Models shows the same catalogue with what each model is good at.

Can I see which model answered?

Yes. Read the X-Flux-Model response header to see which model served the request, and X-Flux-Routed to see whether Flux changed your requested model. When you pin, X-Flux-Routed should be false. See Receipts: which model, what cost.