per-model-breakdown-analytics · EN · 2026-10-17

Which Model Ate Your Budget: Reading the Per-Model Breakdown on One Key

Learn how to read the per-model breakdown on a single API key to identify which model is consuming your budget. This guide explains how to split spend and token counts by model, interpret the data, and take action to optimize costs in a multi-model workflow.

Why Per-Model Breakdown Matters

When you use a single API key to access multiple models (e.g., Claude, GPT, DeepSeek, Qwen, GLM, Kimi), your total spend can obscure which model is driving costs. A per-model breakdown reveals the expensive backend, enabling targeted optimizations.

Accessing the Breakdown

Most API aggregators provide usage dashboards or API endpoints that return per-model statistics. Look for:

  • A usage page in your dashboard with filters for date range and API key.
  • An API endpoint (e.g., /v1/usage) that returns JSON with per-model token counts and costs.
  • Export options (CSV) for deeper analysis.

If your provider doesn't expose this, you can instrument your own logging by recording model name and token usage per request.

Reading the Metrics

Key metrics to extract per model:

  • Input tokens: Tokens sent in prompts.
  • Output tokens: Tokens generated.
  • Total tokens: Sum of input and output.
  • Cost: Calculated as (official price × 1.3) for users. Contributors see credits at (official price × 1.1) or (official price × 1.2) for premium models.

Compare these across models to spot outliers.

Identifying the Expensive Backend

  1. Sort by cost: The model with the highest total cost is your primary budget consumer.
  2. Check token efficiency: High token counts with low output may indicate verbose prompts or inefficient model behavior.
  3. Normalize by request: Calculate average tokens per request to see if a model is used more heavily or with larger payloads.
  4. Compare to official pricing: Since you pay official price × 1.3, the relative cost between models reflects their official rates. A model with high official rates will amplify your spend.

Example Analysis

Suppose your breakdown shows:

  • Claude: 50,000 input tokens, 10,000 output tokens, cost $X.
  • GPT: 100,000 input tokens, 20,000 output tokens, cost $Y.
  • DeepSeek: 200,000 input tokens, 40,000 output tokens, cost $Z.

If Z is disproportionately high, DeepSeek is the expensive backend despite lower official rates—likely due to higher volume. Conversely, if Claude's cost is high despite lower tokens, its higher official rate is the driver.

Taking Action

  • Route wisely: Send simple tasks to cheaper models (e.g., DeepSeek, Qwen) and complex tasks to premium models.
  • Optimize prompts: Reduce input tokens by trimming context or using summarization.
  • Set budgets: Use per-model spending limits if your provider supports them.
  • Monitor regularly: Revisit the breakdown weekly or after workflow changes.

Conclusion

A per-model breakdown on your API key is a powerful tool to uncover hidden costs. By splitting spend and token counts by model, you can pinpoint the expensive backend and make informed decisions to optimize your multi-model workflow.