Skip to content

PRICING.md

Latest commit

 

History

History
137 lines (103 loc) · 3.91 KB

PRICING.md

File metadata and controls

137 lines (103 loc) · 3.91 KB

Pricing Feature

The package now includes automatic cost calculation for all API requests based on actual provider pricing.

Features

  • Automatic Cost Calculation: Costs are calculated automatically for every request
  • Separate Input/Output Costs: Track input tokens, output tokens, and total costs separately
  • Up-to-date Pricing: Includes pricing data from multiple providers (last updated: 2025-10-20)
  • Injectable Pricing: Update pricing data externally when needed

Cost Fields in Usage

Every LLMResponse includes a Usage object with the following cost fields:

response.usage.input_cost_usd     # Cost for input tokens (in USD)
response.usage.output_cost_usd    # Cost for output tokens (in USD)
response.usage.estimated_cost_usd # Total cost (in USD)

Example Usage

Basic Usage (Automatic)

from ai_client import create_ai_client

client = create_ai_client('openai', api_key='your-key')
response = client.prompt('gpt-4o', 'Hello!')

# Access cost information
print(f"Input cost: ${response.usage.input_cost_usd:.6f}")
print(f"Output cost: ${response.usage.output_cost_usd:.6f}")
print(f"Total cost: ${response.usage.estimated_cost_usd:.6f}")

Output:

Input cost: $0.000005
Output cost: $0.000010
Total cost: $0.000015

Updating Pricing Data

When the package pricing data becomes outdated, you can inject your own pricing file:

from ai_client import set_pricing_file, create_ai_client

# Set your custom pricing file
set_pricing_file('/path/to/your/pricing.json')

# Now all requests use your updated pricing
client = create_ai_client('openai', api_key='your-key')
response = client.prompt('gpt-4o', 'Hello!')

Pricing JSON Format

The pricing data follows this format:

{
  "metadata": {
    "created": "2025-09-29T14:20:54.795632",
    "version": "1.3",
    "last_updated": "2025-10-20T00:00:00.000000"
  },
  "pricing": {
    "2025-10-20": {
      "openai": {
        "gpt-4o": {
          "input_price": 2.5,
          "output_price": 10.0,
          "source_url": "...",
          "added": "2025-09-29T21:09:07.468231"
        }
      }
    }
  }
}

Prices are per million tokens (e.g., input_price: 2.5 means $2.50 per 1M input tokens).

Supported Providers

Pricing data is included for:

  • OpenAI: gpt-4o, gpt-4o-mini, gpt-5, gpt-5-mini, etc.
  • Anthropic Claude: claude-3-5-sonnet, claude-3-5-haiku, claude-opus-4, etc.
  • Google Gemini: gemini-2.0-flash, gemini-2.5-pro, gemini-2.5-flash, etc.
  • Mistral: mistral-large, mistral-medium, pixtral-large, etc.
  • OpenRouter: Various models (qwen, llama, grok, etc.)
  • sciCORE: Academic/research models

Cost Calculation Details

  1. The pricing manager loads pricing data from ai_client/pricing.json
  2. For each request, it looks up the model's pricing (most recent date first)
  3. Calculates:
    • input_cost = (input_tokens / 1,000,000) * input_price_per_million
    • output_cost = (output_tokens / 1,000,000) * output_price_per_million
    • total_cost = input_cost + output_cost

When Pricing is Not Available

If pricing data is not available for a model:

  • All cost fields will be None
  • Token counts are still tracked
  • No errors are raised

You can check availability:

if response.usage.estimated_cost_usd is not None:
    print(f"Cost: ${response.usage.estimated_cost_usd:.6f}")
else:
    print("Pricing not available for this model")

Updating the Package Pricing

To update the built-in pricing data:

  1. Update ai_client/pricing.json with new prices
  2. Increment the version in metadata.version
  3. Update metadata.last_updated
  4. Rebuild and redistribute the package

Notes

  • Prices are sourced from official provider documentation
  • Archive.org URLs are included for verification
  • Costs are estimates and may vary slightly from actual billing
  • Some providers (like OpenRouter) include actual costs in their API response, which are used directly