Auto-Routing

When you set model: "auto", the system automatically selects the best model for the task based on complexity, capabilities required, and cost optimization.

How It Works

1. Analyze the request context (context.task, message content, multimodal attachments)
2. Select the optimal model tier (economical, balanced, premium, flagship)
3. Choose the best available model within that tier
4. Apply fallback chain if the primary model fails

Task Types

Specify context.task in your request to optimize model selection:

Value	Description	Recommended Tier
chat_general	General conversations, explanations	economical
write	Content creation (email, post, proposal)	balanced
rewrite_edit	Revision, tone adjustment	economical
extract_structure	Data extraction, JSON/table formatting	economical
reasoning_analysis	Complex analysis, multi-step reasoning	flagship
code	Code generation, debugging, refactoring	premium
data_query	SQL queries, reports, dashboards	premium
research	Research and synthesis	premium

Validation: Invalid task types return HTTP 400 with the list of valid values.

Model Tiers

Tier	Use Case	Typical Models
economical	Simple tasks, low latency	Haiku 4.5, GPT-4.1-nano, Gemini 2.5 Flash Lite
balanced	General purpose	Sonnet 4.5, GPT-4.1-mini
premium	High quality	GPT-4.1, Gemini 2.5 Flash, o3-mini
flagship	Maximum capability	Claude Opus 4.6, o3, Gemini 2.5 Pro

Tier Eligibility by Plan

Auto-routing only selects models within the tiers allowed by your subscription plan:

Plan	Allowed Tiers
free	economical
basic	economical, premium
pro	economical, premium, flagship
enterprise	all

If your plan restricts certain tiers, auto-routing will only consider eligible models. For example, a basic plan using model: "auto" will never select a flagship model.

tip

To access higher-tier models, upgrade your subscription plan.

Routing Response

When auto-routing is used, the response includes routing metadata:

{
  "routing": {
    "is_auto_routed": true,
    "model_chosen": "openai.gpt-4.1-nano",
    "confidence": 0.85
  }
}

Field	Description
is_auto_routed	Whether auto-routing was used
model_chosen	Full model ID selected
confidence	Routing confidence score (0-1)

Multimodal Auto-Routing

When the request contains images or other multimodal content, the router automatically filters to vision-capable models:

{
  "model": "auto",
  "messages": [{
    "role": "user",
    "content": [
      {"type": "text", "text": "Describe this image."},
      {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
    ]
  }],
  "context": { "task": "extract_structure" }
}

---

GET /api/llm/recommendations/:task

Get model recommendations for a specific task type.

Authentication: Not required

Cache: 10 minutes

Parameters:

Param	Type	Description
task	string	Task type (see table above)

Response:

{
  "success": true,
  "task": "code_generation",
  "recommendations": [
    {
      "id": "grok.grok-4-1-fast-reasoning",
      "name": "Grok 4.1 Fast Reasoning",
      "provider": "grok",
      "tier": "flagship",
      "context": "2000000"
    }
  ],
  "count": 5
}