AI Models and OpenRouter Setup

def estimate_cost(num_documents, pages_per_doc, avg_words_per_page, model="openai/gpt-4o-mini"):
    """
    Estimate OpenRouter API cost for batch processing
    
    Pricing (per 1M input tokens):
    - gpt-4o-mini: $0.15
    - gemini-flash-1.5: $0.075
    - claude-3-haiku: $0.25
    - claude-3.5-sonnet: $3.00
    """
    # Rough estimate: 750 words = 1000 tokens
    tokens_per_doc = (pages_per_doc * avg_words_per_page) / 0.75
    total_tokens = num_documents * tokens_per_doc
    
    # Model pricing per 1M tokens (input)
    prices = {
        "openai/gpt-4o-mini": 0.15,
        "google/gemini-flash-1.5": 0.075,
        "anthropic/claude-3-haiku": 0.25,
        "anthropic/claude-3.5-sonnet": 3.00
    }
    
    cost_per_million = prices.get(model, 0.15)
    total_cost = (total_tokens / 1_000_000) * cost_per_million
    
    return {
        "documents": num_documents,
        "total_tokens": int(total_tokens),
        "estimated_cost": round(total_cost, 4),
        "cost_per_document": round(total_cost / num_documents, 6)
    }

# Example: 1000 documents, 3 pages each, 300 words/page
estimate = estimate_cost(
    num_documents=1000,
    pages_per_doc=3,
    avg_words_per_page=300,
    model="openai/gpt-4o-mini"
)

print(f"Total cost: ${estimate['estimated_cost']}")
print(f"Cost per document: ${estimate['cost_per_document']}")
# Output:
# Total cost: $0.18
# Cost per document: $0.00018

# OpenRouter endpoint
BASE_URL = "https://openrouter.ai/api/v1"

# Timeouts
CONNECT_TIMEOUT = 10  # seconds to establish connection
READ_TIMEOUT = 120    # seconds to wait for response

# Retry settings
MAX_RETRIES = 3       # retry failed requests
RETRY_DELAY = 2       # initial delay between retries

# Rate limiting
RETRY_DELAY_MULTIPLIER = 1.5  # exponential backoff
MAX_DELAY_BETWEEN_REQUESTS = 120  # cap delay at 2 minutes

{
  "model": "openai/gpt-4o-mini",
  "messages": [
    {
      "role": "system",
      "content": "You are a document search-tagging expert..."
    },
    {
      "role": "user",
      "content": "Analyze the document below and return exactly 8 search tags..."
    }
  ],
  "max_tokens": 700,
  "temperature": 0.2
}