Nitro Variant

:nitro asks for speed. It sorts the model's providers by throughput — tokens per second — and lets a provider's priority tier compete too.

Usage#

curl https://opendunes.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.6-luna:nitro",
    "messages": [{ "role": "user", "content": "Reply to this customer in one line: where is my parcel?" }]
  }'

What it does#

1. Sorts every eligible endpoint by throughput — the same as provider.sort: "throughput".
2. Admits priority-tier endpoints to that sort. They get no special treatment: a priority endpoint serves only when it is the fastest.

The model is the same model — same weights, same context. Only the provider that serves it changes.

Billing#

You pay the endpoint that served, in DA. A priority endpoint charges its priority price; a standard one, its standard price. For the speed sort without priority tiers, set provider.sort: "throughput" instead.

Combining variants#

Among :nitro, :floor and :exacto, the last one named wins.