Cost Tracking

Proxle automatically calculates the cost of every LLM request, giving you visibility into where your money goes.

How It Works

  1. After each request, Proxle extracts token counts from the response
  2. Token counts are matched against the provider pricing table for the model used
  3. Cost is calculated as: (input_tokens * input_price + output_tokens * output_price) / 1,000,000
  4. The cost and token counts are stored alongside the request

Confidence Levels

| Level | Description | |-------|-------------| | Exact | Token counts came from the provider's response (e.g., OpenAI usage field) | | Approximate | Token counts were estimated using tiktoken or character-based approximation |

Most OpenAI and Anthropic requests have exact confidence. Cohere and Gemini may use approximations depending on the response format.

Cost Attribution by Feature

Use the metadata.feature field in your SDK calls to attribute costs to features:

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[...],
    metadata={"feature": "chat_assistant"}
)

Then view the breakdown in Dashboard > Analytics > Cost by Feature.

Cost Attribution by User

Use the metadata.user_id field to track per-user costs:

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[...],
    metadata={"user_id": "user_123"}
)

View per-user costs in Dashboard > Analytics > Cost by User.

Analytics Dashboard

The analytics page shows:

  • Total cost over a time period (7, 14, or 30 days)
  • Cost by provider - Which providers cost the most
  • Cost by model - Which models cost the most
  • Cost by feature - Which features are most expensive
  • Cost by user - Which users generate the most cost
  • Cost over time - Daily cost trends
  • Cache savings - Money saved through cache hits

Supported Models and Pricing

Proxle maintains a pricing table for all supported models:

OpenAI

  • GPT-4o, GPT-4o-mini, GPT-4-turbo, GPT-4, GPT-3.5-turbo
  • O1, O1-mini, O3-mini

Anthropic

  • Claude Opus 4, Claude Sonnet 4, Claude 3.5 Sonnet, Claude 3.5 Haiku, Claude 3 Opus, Claude 3 Haiku

Cohere

  • Command R+, Command R, Command, Command Light

Google Gemini

  • Gemini 2.0 Flash, Gemini 1.5 Pro, Gemini 1.5 Flash, Gemini 1.0 Pro

Azure OpenAI

  • GPT-4o, GPT-4o-mini, GPT-4-turbo, GPT-4, GPT-3.5-turbo

Pricing is updated regularly to reflect provider price changes.

Smart CachingRequest Replay