​
Model Selection in Routstr

When using Routstr’s API, you have several options for selecting which AI model to use for your requests. This guide explains the available options and best practices.

​
Specifying a Model

You can directly specify which model to use in your API requests:

curl -X POST https://api.routstr.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cashuA1DkpMbgQ9VkL6U..." \
  -d '{
    "model": "gpt-4",
    "messages": [
      {
        "role": "user", 
        "content": "Explain quantum computing"
      }
    ]
  }'

Common model options include:

  • gpt-4 - OpenAI’s GPT-4 model
  • llama-3-70b-chat - Meta’s Llama 3 (70B parameter) chat model
  • claude-3-opus - Anthropic’s Claude 3 Opus model

​
Using the Default Route

If you don’t specify a model, Routstr will route your request to an appropriate model based on your request content and model availability:

curl -X POST https://api.routstr.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cashuA1DkpMbgQ9VkL6U..." \
  -d '{
    "messages": [
      {
        "role": "user", 
        "content": "Explain quantum computing"
      }
    ]
  }'

The default routing considers:

  1. Availability: Which models are currently accessible and responding quickly
  2. Cost: Models with lower token prices may be preferred
  3. Capability: The complexity of your request may route to more powerful models

​
Choosing Models Based on Capabilities

Different models have different capabilities. For specialized tasks, use the appropriate model:

​
For Text Generation

curl -X POST https://api.routstr.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cashuA1DkpMbgQ9VkL6U..." \
  -d '{
    "model": "llama-3-8b-chat",
    "messages": [
      {
        "role": "user", 
        "content": "Write a short poem about autumn"
      }
    ]
  }'

​
For Code Generation

curl -X POST https://api.routstr.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cashuA1DkpMbgQ9VkL6U..." \
  -d '{
    "model": "gpt-4",
    "messages": [
      {
        "role": "user", 
        "content": "Write a Python function to calculate Fibonacci numbers"
      }
    ]
  }'

​
For Image Understanding (Vision)

curl -X POST https://api.routstr.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cashuA1DkpMbgQ9VkL6U..." \
  -d '{
    "model": "gpt-4-vision",
    "messages": [
      {
        "role": "user", 
        "content": [
          {
            "type": "text",
            "text": "What's in this image?"
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://example.com/image.jpg"
            }
          }
        ]
      }
    ]
  }'

​
Model Parameters

You can customize how a model generates responses by adjusting parameters:

curl -X POST https://api.routstr.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cashuA1DkpMbgQ9VkL6U..." \
  -d '{
    "model": "gpt-4",
    "temperature": 0.7,
    "max_tokens": 500,
    "top_p": 0.95,
    "messages": [
      {
        "role": "user", 
        "content": "Explain quantum computing"
      }
    ]
  }'

Common parameters include:

  • temperature: Controls randomness (0.0 to 1.0, lower is more deterministic)
  • max_tokens: Maximum length of the generated response
  • top_p: Nucleus sampling parameter (0.0 to 1.0)
  • top_k: Limits vocabulary to top k options (only for some models)
  • presence_penalty: Reduces repetition of previously mentioned topics
  • frequency_penalty: Reduces repetition of frequently used tokens

​
Model Pricing

Different models have different pricing per token:

ModelInput Price (per 1K tokens)Output Price (per 1K tokens)
gpt-3.5-turbo$0.0005$0.0015
gpt-4$0.03$0.06
claude-3-opus$0.015$0.075
llama-3-70b$0.0007$0.0014

Consider the balance between model capability and price for your use case.

​
Using the Python OpenAI Client

If you’re using Python, you can use the OpenAI client to select models:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.routstr.com/v1",
    api_key="cashuA1DkpMbgQ9VkL6U..."
)

# Specify a model
completion = client.chat.completions.create(
    model="gpt-4",
    messages=[
        {"role": "user", "content": "Write a short story"}
    ],
    temperature=0.7,
    max_tokens=2000
)

print(completion.choices[0].message.content)

​
Best Practices

  1. Start with a general-purpose model like gpt-3.5-turbo for basic tasks
  2. Upgrade to more powerful models for complex reasoning or specialized tasks
  3. Use vision models when working with images
  4. Balance cost and capability - only use expensive models when needed
  5. Be specific in your prompts about quality and format requirements

​
Next Steps