model-usage

Per-model cost summaries from CodexBar logs are computed by parsing daily modelBreakdowns and modelsUsed data structures to aggregate total costs per model. The process selects the latest date entries to output either a human-readable summary or a structured JSON object detailing model-level expenses.

The best way to analyze model usage costs for Codex and Claude is parsing local cost logs to extract per-model expense data. This method aggregates usage datasets to provide a full breakdown by model or identify the current top-cost model for budgeting and auditing tasks.

You can output cost analysis results as a JSON object containing total costs per model. The computation parses local log data and supports flexible output formats, allowing generation of either a human-readable text summary or a structured JSON payload for reporting.

The per-model usage breakdown requires a JSON-like structure containing daily modelBreakdowns and modelsUsed fields. The Skill reads directly from local CodexBar logs formatted with these fields, or accepts a provided JSON payload matching this structure to compute accurate cost summaries.

The cost breakdown shows the latest date because the computation logic automatically selects the most recent date entries from the modelBreakdowns data when summarizing. This selection ensures the summary reflects current model usage and expenses without duplicating historical daily aggregates.