Codex Cost Dashboard

Local Codex cost estimator. It scans Codex Desktop JSONL logs under ~/.codex, extracts token usage snapshots, and estimates USD cost from prices.json.

This is an estimate, not an official bill. Codex subscription accounting may differ from API list pricing, and local logs do not expose every possible plan-specific adjustment. The useful part is relative visibility: which sessions, workspaces, or days are consuming the most tokens.

Privacy

This tool runs locally and reads local Codex session logs. Those logs can contain prompts, workspace paths, model names, and timing metadata. The generated dashboard.html and CSV reports are ignored by git by default because they may include private local usage details.

Review any screenshots, exported reports, or generated dashboards before sharing them publicly.

Usage

python3 scan_codex_cost.py --today
python3 scan_codex_cost.py --since 2026-05-01 --group-by session
python3 scan_codex_cost.py --since 2026-05-01 --group-by turn --limit 30
python3 scan_codex_cost.py --cwd /path/to/workspace --group-by day
python3 scan_codex_cost.py --today --export-csv reports/codex-cost-today.csv

Dashboard

Generate a self-contained local HTML dashboard:

python3 generate_dashboard.py
open dashboard.html

By default this includes records from the first day of the current month through now. To change the period:

python3 generate_dashboard.py --since 2026-05-01
python3 generate_dashboard.py --since 2026-05-01 --fallback-model gpt-5.3-codex

The dashboard embeds a data snapshot. Re-run generate_dashboard.py whenever you want to refresh it.

For live updates, run the local server instead:

python3 serve_dashboard.py --open

Then leave the terminal running. The browser page calls /api/data every 30 seconds and keeps your current filters while refreshing the charts and table. The session table defaults to newest-opened sessions first, with cost, token, and turn-count sorting still available. Stop it with Ctrl-C.

Fields

• input_tokens: total input tokens from local Codex logs.
• cached_input_tokens: subset of input tokens logged as cached.
• billable_input_tokens: input_tokens - cached_input_tokens.
• output_tokens: output tokens from local Codex logs. Reasoning tokens are already included in output_tokens in the observed Codex log format.
• estimated_usd: local estimate using prices.json.

Pricing

prices.json was seeded from the OpenAI API pricing page checked on 2026-05-15: https://openai.com/api/pricing/.

Update prices.json when model pricing changes or when you want to use a custom internal rate.

gpt-5.3-codex-spark is intentionally not priced in the default table because the OpenAI Codex rate card describes it as a research preview with non-final rates. Use --fallback-model gpt-5.3-codex if you want a rough placeholder.