Every row at the lowest eligible price.

Pipeline rows are ordinary requests: each one is ranked by cost across every eligible endpoint and returns its own charge. Key budgets stop a runaway job, and exports make unit economics visible before the invoice.

Start with this policy Read the guide

WHAT YOU GET

Cost-ranked routing on every row

Budgets that warn and hard-stop per key or project

Exports: per-request route, usage and charge

MODELS FOR THIS WORKLOAD · LIVE CATALOG

Chat & reasoning Embeddings Image generation

batch.py · OpenAI SDK, base URL changed

total = 0
for row in rows:
    res = client.chat.completions.create(
        model="deepseek/deepseek-v4-flash-0731",
        messages=prompt_for(row),
    )
    total += res.yelno["charge"]["nanos"]   # integer nanos · final

# the console exports every request's route,
# usage and charge for your warehouse

Go live in three steps

Pick the model by price

Sort the catalog by published price and pick the cheapest model that passes your evals.

Give the job a key budget

Warn at a threshold, hard stop at the limit; rows already answered keep their results.

Export the ledger

Per-request route, usage and charge for your finance pipeline.

Batch & pipelines guideproduction in four steps

1. Model

Filter the catalog by type and published price; run your evals on the cheapest candidates.

2. Budget

Give the pipeline its own key with a warning threshold and a hard stop.

3. Fallback

When an endpoint rate-limits, the gateway moves the row to the next eligible endpoint inside your constraints.

4. Ledger

Export per-request route, usage and charge to your warehouse.

guide copy · the code calls the gateway's OpenAI-compatible routes; model slugs are examples from the catalog · not yet connected