Models

One API for model discovery and capability metadata

TheRouter.ai publishes model metadata through a standardized Models API. Use it to find supported modalities, context lengths, pricing, and parameters before sending traffic.

Model naming convention

Model IDs use a consistent provider/model-name format, so your routing and analytics stay portable across providers.

model-ids.txt

anthropic/claude-sonnet-4.5
openai/gpt-4o
google/gemini-2.5-pro
mistralai/mistral-large

Listing models

Query /models to retrieve all available models and metadata. You can also filter on the web UI and keep systems synchronized using the RSS endpoint.

cURL

curl https://api.therouter.ai/v1/models \
  -H "Authorization: Bearer $THEROUTER_API_KEY"

Core schema

Each model object includes identity fields, context limits, architecture capabilities, and pricing data.

model-object.json

{
  "id": "google/gemini-2.5-pro",
  "name": "Gemini 2.5 Pro",
  "context_length": 1048576,
  "architecture": {
    "input_modalities": ["text", "image", "audio", "video", "file"],
    "output_modalities": ["text"],
    "tokenizer": "sentencepiece",
    "instruct_type": null
  },
  "pricing": {
    "prompt": "0.00000125",
    "completion": "0.00000500",
    "request": "0",
    "image": "0",
    "web_search": "0"
  },
  "supported_parameters": ["temperature", "top_p", "tools", "response_format"]
}

Pricing units

Pricing fields are stored in USD per token or per unit. For human-readable display, convert to per-million token rates.

pricing-display.txt

prompt: 0.00000125 USD/token  ->  $1.25 per 1M input tokens
completion: 0.00000500 USD/token ->  $5.00 per 1M output tokens

Tokenizer differences

Token counts vary by model tokenizer, so the same prompt can produce different total usage and cost across providers.

Context lengths and provider limits

Use context_length and provider metadata to preflight request sizes and avoid token overflow failures in production.

Deploy-time validation

Validate model capabilities and supported parameters at startup so incompatible settings fail early.