Nitro Variant

Not yet available.

The :nitro variant routes your request to provider instances reserved for high-throughput workloads — dedicated capacity with lower queuing and faster token generation. Use it when speed matters more than cost: real-time applications, latency-sensitive pipelines, or interactive experiences where a slow response breaks the user experience.

Usage#

curl https://opendunes.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-llama/llama-4-maverick:nitro",
    "messages": [{ "role": "user", "content": "Write a product description in under 5 seconds." }]
  }'

What nitro changes#

Appending :nitro to a model slug tells OpenDunes to route exclusively to providers and instances flagged for high throughput. In practice this means:

• Lower queuing — requests bypass the shared queue and land on reserved capacity.
• Higher tokens/second — provider instances are typically larger or more lightly loaded.
• Higher cost — the premium capacity costs more per token in DA.

The model itself is unchanged — you get the same weights, capabilities, and context limits as the standard variant. Only the infrastructure running it is different.

When to use nitro#

Pricing (DA)#

Nitro requests are billed at a higher per-token DA rate than the standard variant. The premium varies by model — check the catalog:

GET /api/models/meta-llama/llama-4-maverick

The response includes pricing.nitro as a separate input/output rate in DA per million tokens. For cost-sensitive workloads, compare the nitro premium against the value of the latency improvement before adopting it at scale.

Throughput guarantees#

The :nitro variant prioritizes your request, but does not guarantee a specific tokens/second SLA — provider capacity fluctuates. For strict SLA requirements, contact support about reserved capacity arrangements.

Using with the OpenAI SDK#

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://opendunes.com/api/v1",
    api_key=os.environ["OPENDUNES_API_KEY"],
)

resp = client.chat.completions.create(
    model="meta-llama/llama-4-maverick:nitro",
    messages=[{"role": "user", "content": "List 5 startup ideas in Algeria."}],
    stream=True,
)
for chunk in resp:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

Note. Combining :nitro with other variants (e.g., model:nitro:thinking) is not supported. Each request uses one variant suffix. If you need both high throughput and extended reasoning, contact support.