LiteLLM Proxy OpenRouter Price Updater

A CLI tool that compares local OpenRouter model pricing in LiteLLM proxy configuration files with the actual pricing from OpenRouter's API. This helps maintain accurate cost tracking by identifying missing or outdated pricing information.

This script assumes that in your openrouter preferences you set Default Provider Sort to Price (cheapest first).

You might also be interested in my other script: OpenRouter to Langfuse Model Pricing Sync.

Features

• Fetches current pricing data from OpenRouter API
• Compares local configuration pricing with API pricing
• Identifies missing pricing fields in your configuration
• Detects price mismatches between local config and API
• Supports all OpenRouter pricing fields:
  • Input cost per token (prompt pricing)
  • Output cost per token (completion pricing)
  • Cache creation input token cost
  • Cache read input token cost
  • Input cost per image
  • Output cost per reasoning token
• Detailed logging with color-coded output
• Exit codes for CI/CD integration

Installation

Prerequisites

• Python 3.7+
• pip

Install Dependencies

pip install -r requirements.txt

Usage

Basic Usage

python litellm_proxy_openrouter_price_updater.py --config path/to/your/litellm_config.yaml

Example Output

2024-01-15 10:30:12.345 | INFO     | Fetched 150 models from OpenRouter API
2024-01-15 10:30:12.456 | INFO     | Found 5 OpenRouter models in config
2024-01-15 10:30:12.567 | INFO     | Checking model: GPT-4 Turbo (openrouter/openai/gpt-4-turbo)
2024-01-15 10:30:12.678 | SUCCESS  | Pricing is up to date for GPT-4 Turbo
2024-01-15 10:30:12.789 | WARNING  | Pricing discrepancies for Claude 3:
2024-01-15 10:30:12.890 |          |   - Missing cache_read_input_token_cost: API has input_cache_read=0.000001
2024-01-15 10:30:12.991 |          |   - output_cost_per_token mismatch: local=0.000015, API=0.000016
2024-01-15 10:30:13.100 | INFO     | Pricing warnings for Mixtral:
2024-01-15 10:30:13.200 |          |   - API has pricing for 'web_search' (0.003) - not tracked by LiteLLM
2024-01-15 10:30:13.300 | INFO     | ============================================================
2024-01-15 10:30:13.400 | INFO     | PRICING CHECK RECAP
2024-01-15 10:30:13.500 | INFO     | ============================================================
2024-01-15 10:30:13.600 | INFO     | Total models checked: 5
2024-01-15 10:30:13.700 | INFO     | Models with pricing issues: 1
2024-01-15 10:30:13.800 | INFO     | Models with warnings: 1
2024-01-15 10:30:13.900 | INFO     | Total pricing issues: 2
2024-01-15 10:31:14.000 | INFO     | Total warnings: 1

Configuration File Format

The tool expects a LiteLLM proxy configuration file with OpenRouter models defined like this:

model_list:
  - model_name: "gpt-4-turbo"
    litellm_params:
      model: "openrouter/openai/gpt-4-turbo"
      input_cost_per_token: 0.00001
      output_cost_per_token: 0.00003
      cache_creation_input_token_cost: 0.0000125
      cache_read_input_token_cost: 0.0000025
      input_cost_per_image: 0.001445
      output_cost_per_reasoning_token: 0.00006

Exit Codes

• 0: All pricing is up to date (warnings are allowed)
• 1: Pricing discrepancies found or error occurred

This makes the tool suitable for use in CI/CD pipelines to catch pricing configuration drift. Note that informational warnings (like unsupported API pricing fields) do not cause a non-zero exit code.

Development

Error Handling

The tool handles various error conditions gracefully:

• Missing config file: Clear error message with file path
• Invalid YAML: Detailed parsing error information
• API failures: Network timeout and HTTP error handling
• Missing models: Warning when local models aren't found in API
• Model modifiers: Automatically strips modifiers (e.g., :nitro) from model IDs when comparing with API
• Unsupported pricing fields: Some API pricing fields (like web_search) generate informational warnings rather than errors

Output Types

The tool distinguishes between two types of issues:

• Discrepancies: Missing or mismatched pricing that should be fixed (causes exit code 1)
• Warnings: Informational messages about API pricing fields not tracked by LiteLLM (exit code 0)

At the end of each run, a detailed recap summary shows:

• Total models checked
• Models with pricing issues vs warnings
• Counts of total issues and warnings
• Lists of affected models

Pricing Fields Supported

Local Config Field	OpenRouter API Field	Description
input_cost_per_token	prompt	Cost per input token
output_cost_per_token	completion	Cost per output token
cache_creation_input_token_cost	input_cache_write	Cost to write to cache
cache_read_input_token_cost	input_cache_read	Cost to read from cache
input_cost_per_image	image	Cost per input image
output_cost_per_reasoning_token	internal_reasoning	Cost per reasoning token

Informational Fields

Some OpenRouter API pricing fields generate warnings but don't require configuration updates:

OpenRouter API Field	Status	Description
web_search	Warning only	Web search cost - not tracked by LiteLLM

License

This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.

Acknowledgments

• Generated with assistance from aider.chat
• Uses the OpenRouter API for pricing data
• Built for use with LiteLLM Proxy