Auto Router

Not yet available.

The Auto Router removes the model selection decision from your code entirely. Set model to "opendunes/auto" and OpenDunes analyzes the prompt, infers the task type, and routes to the best available model — balancing accuracy, cost, and latency in real time.

Usage#

curl https://opendunes.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "opendunes/auto",
    "messages": [{ "role": "user", "content": "Write a Redis Lua script that atomically decrements a balance." }]
  }'

The model field in the response tells you which model was chosen. Every other field in the response is identical to a direct model request.

How selection works#

The Auto Router runs a fast classification step on the prompt before dispatching it. It considers:

1. Task type — coding, reasoning, writing, instruction-following, multilingual, etc.
2. Prompt length — long prompts are routed away from models with shorter context windows.
3. Cost tier — the router applies a default cost tier (configurable) to filter out models above a DA threshold.
4. Provider availability — unhealthy or rate-limited providers are excluded from the candidate set.

The classification step adds less than 100ms of overhead in most cases.

Cost tier#

By default, the Auto Router picks the best model across all cost tiers. To constrain the selection to a budget, pass max_cost_da per request:

{
  "model": "opendunes/auto",
  "max_cost_da": 0.3,
  "messages": [{ "role": "user", "content": "What is photosynthesis?" }]
}

You can also set a default cost ceiling for all Auto Router requests in your dashboard under Settings → Routing defaults.

Transparency#

Because the router selects the model dynamically, the DA cost per request varies. The model field in the response always tells you what ran. The usage field gives you the exact token counts, which you can cross-reference with the catalog's DA pricing to understand the cost.

The router does not currently expose a "why did you pick this" field — the selected model is the only output. This may change.

Opting out of a model#

If there are models you never want the Auto Router to choose — for compliance, budget, or quality reasons — add them to your account's router exclusion list in the dashboard. Excluded models are skipped regardless of classification result.

Pricing (DA)#

You pay the selected model's standard per-token DA rate. The Auto Router itself has no additional charge. Requests routed to cheap models cost less; requests routed to frontier models cost more. Use max_cost_da to keep variance predictable.

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="opendunes/auto",
    messages=[{"role": "user", "content": "Translate 'invoice total' to Arabic and French."}],
)
print(resp.choices[0].message.content)
print("Routed to:", resp.model)

Note. The Auto Router is designed for general workloads. For accuracy-critical tasks, use Auto Exacto. For cost-critical workloads, use the Free Models Router.