Auto Mode
Set "model": "auto" and DeepNexus picks a model for each request based on what you send. You get flagship quality where it matters and low-cost speed everywhere else, without changing code.
Routing rules
The router looks at the latest user message:
| Request | Routed to | Why |
|---|---|---|
| Contains an image | mimo-v2.6-pro | Vision-capable flagship |
Contains code (code fences, function, class, def, import, SQL, HTML tags) | deepseek-v4-pro | Strongest reasoning and coding |
| Longer than 2,000 characters | deepseek-v4-pro | Long-context analysis |
| Everything else | deepseek-v4.1-flash | Fast, cheap, strong general model |
The response's model field tells you which model served the request, and the Usage page bills it at that model's price.
Example
curl http://localhost:3000/v1/chat/completions \
-H "Authorization: Bearer $DEEPNEXUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "auto", "messages": [{"role": "user", "content": "def fib(n): ... make this iterative"}]}'
{ "model": "deepseek-v4-pro", "choices": [ ... ] }
In Chat
Choose Auto in the model selector. The model that answered is shown under each reply.
When not to use Auto
Pin a specific model when you need reproducible outputs across requests, a particular context window, or a feature only one model supports.