ryoppippi/ccusage v0.4.0 on GitHub

🚀 Major Changes

Multiple Cost Calculation Modes

In response to issue #4 regarding cost calculation discrepancies and missing costUSD fields in newer Claude Code versions, this release introduces flexible cost calculation modes:

auto (default): Intelligently uses pre-calculated costUSD when available, otherwise calculates from token counts
calculate: Always calculates costs from token counts using LiteLLM pricing data
display: Only shows pre-calculated costUSD values, displays 0 for missing costs

# Examples
npx ccusage@latest --mode auto      # Smart mode - best of both worlds
npx ccusage@latest --mode calculate # Force calculation from tokens
npx ccusage@latest --mode display   # Only show stored costs

Internal Debug Functionality

Added internal debug module for cost verification (not exposed as CLI command):

Compares pre-calculated vs token-calculated costs
Reports mismatches exceeding 10% threshold
Helps developers verify cost accuracy across different Claude Code versions
This is an experimental feature used for development and testing

npx ccusage@latest --debug

🎯 Key Features

Cost Calculation Improvements

Integration with LiteLLM's pricing database for accurate model costs
Better handling of missing cost data

Technical Enhancements

Type-safe enum validation for mode arguments
Improved error handling and edge case coverage
Enhanced test coverage with real Claude 4 models
Better separation of concerns in cost calculation logic

📝 Documentation Updates

Added comprehensive tips to CLAUDE.md
Documented all cost calculation modes
Added important disclaimers about virtual cost calculations

💡 Background

This release addresses the community's need for flexible cost tracking as Claude Code evolves. With some versions not including pre-calculated costs, users can now choose how they want costs to be determined - whether from stored values, calculated from tokens, or a smart combination of both.

As a result of the verification, we found that the costUSD was sometimes half the price of the API price. So we think the apparent cost will be even more inflated with this change.
In any case, it is not possible to measure costs accurately. Just look at it as a reference.

⚠️ Important Notes

Cost calculations are based on LiteLLM's pricing data and may not reflect actual billing
The calculate mode provides estimates based on token counts and model pricing
For official billing information, always refer to your Claude account dashboard

🙏 Acknowledgments

Special thanks to all contributors who shared their logs and helped identify the cost calculation discrepancies in issue #4 ( @nnamon @itsUndefined @mimikun )

🚨 Breaking Changes

Introducing debug command - by @ryoppippi (2b8c5)

🚀 Features

Implement pricing-fetcher - fetch from litellm's data - by @ryoppippi (599b5)
Collect versions per session - by @ryoppippi (75f40)
Add CostMode type - by @ryoppippi (77916)
Cost mode option to args - by @ryoppippi (a6858)
Implement cost mode options to command - by @ryoppippi (089d4)
Extract duplicated process of calculate cost - by @ryoppippi (47cb8)
data-loader: Add loader forr fetching model - by @ryoppippi (4af0b)

🐞 Bug Fixes

Argument type of debug should be enum - by @ryoppippi (78fbd)
Stop fetching pricing when display mode - by @ryoppippi (4f8fd)
debug.ts: Do not use magic number - by @ryoppippi and Copilot (0b26a)
log: Add types to log function - by @ryoppippi (4ed8b)