Skip to subsidy history
aicharts.io

Subsidy for ChatGPT Pro 20x subscription

Daily API-retail-equivalent value of seven complete UTC days from one user's available local Codex logs on one machine, projected over four weeks and compared with a $200 ChatGPT Pro plan-price unit.

31 settled observations · –· Updated

Four-week API-equivalent multiple

307.1×

$15,353 × 4 ÷ $200

$61,413 projected 28-day API-retail-equivalent value

Measured 31-day total: $63,289 from to

Weekly quota status is not observed. The ×4 is a fixed 28-day convention, not four observed exhausted allocations, and the usage cannot be durably attributed to one subscription or billing source.

History

Each point is the preceding seven complete UTC days, priced at model-specific API rates and compared with the $200 plan price.

Four-week API-retail-equivalent multiple over time31 daily observations. The line shows the API-retail-equivalent plan-price multiple derived from each trailing seven-day period. Historical points are subdued and the latest point is emphasized.
Each point covers seven complete UTC daysScroll horizontally for all dates
View all 31 observations
Historical local Codex API-equivalent observations
ObservedPeriodTokensFour-week estimateMultiple
– 34.4B$76,981384.9×
– 35B$77,699388.5×
– 33.3B$73,763368.8×
– 29.3B$64,544322.7×
– 25.1B$54,834274.2×
– 24.3B$52,575262.9×
– 21.1B$45,121225.6×
– 17.1B$36,041180.2×
– 15.5B$32,456162.3×
– 18.1B$38,010190.1×
– 20.2B$42,449212.2×
– 25.4B$52,928264.6×
– 28B$57,193286.0×
– 29.1B$58,957294.8×
– 28.3B$57,079285.4×
– 30.6B$61,078305.4×
– 30.2B$59,504297.5×
– 32.6B$63,628318.1×
– 32.4B$62,080310.4×
– 34.6B$64,239321.2×
– 35.2B$63,592318.0×
– 36.9B$66,925334.6×
– 37.9B$68,775343.9×
– 38.4B$69,840349.2×
– 38.1B$69,236346.2×
– 36.7B$66,864334.3×
– 29.7B$55,361276.8×
– 29.1B$56,302281.5×
– 31.9B$61,864309.3×
– 30.8B$60,266301.3×
– 31.1B$61,413307.1×

Calculation

Available local Codex task logs, including child agents, are deduplicated into daily token buckets and repriced with model-specific API-price estimates from the checked manifest. Each point sums seven settled days, multiplies that value by 4, then divides by the $200 plan price.

The published collector does not track weekly quota exhaustion or reset windows. Its line is realized local usage, not an allowance ledger.

Measurement details, limits, and sources

Cached input is a subset of input and reasoning tokens are a subset of output, so neither is counted twice. The pricing manifest was frozen (SHA-256 911b73c86f8e).

Measurement revision 2026-08-25.2 pins the parser, adapter, rolling-window math, and updater source. A changed measurement manifest requires the retained series to be recomputed before publication.

This is an API-retail-equivalent estimate of one user's available local Codex logs on one machine, not OpenAI's internal serving cost or an audited subsidy, and it is not a platform-wide or representative ChatGPT Pro estimate. The published series does not observe whether a weekly quota was exhausted or when it reset; exact ×4 is a normalization convention, not four observed exhausted allocations. Codex logs do not retain a durable account ID or billing mode, so API-key or otherwise API-billed usage, multiple accounts, subscriptions, purchased ChatGPT credits, free or reset credits, and temporary promotions cannot be distinguished or excluded. The internal codex-auto-review alias uses the explicit GPT-5.6 Luna proxy rate. Non-token product features and fees absent from the logs are excluded.

Limits

  • This is one user's available local logs on one machine, not a platform-wide or representative ChatGPT Pro estimate.
  • The session directory has no durable account ID or billing mode. Historical account switches, multiple subscriptions, API-key or otherwise API-billed usage, purchased ChatGPT credits, free or reset credits, and promotions cannot be separated.
  • The line is realized local usage pace. It does not reconstruct allowance resets, prove that a weekly limit was exhausted, or establish how much usage one subscription supplied.
  • Every point uses seven complete UTC days. Open current-day usage is excluded until the next settled bucket, and the scheduled collector's own small Codex token use appears in the next settled bucket.
  • An unknown recorded model blocks publication. Unrecorded or non-token product operations are excluded. The internal codex-auto-review alias uses a manifest proxy estimate rather than a published first-party API rate.
  • The title's “20x” is the advertised Pro usage tier, not this chart's calculated API-equivalent multiple.

Sources

  1. ChatGPT Pro plan source
  2. https://github.com/hraness/aicharts/blob/main/data/gpt-subsidy-pricing.json
  3. https://developers.openai.com/api/docs/models/gpt-5.6-sol
  4. https://learn.chatgpt.com/docs/pricing
  5. https://github.com/hraness/aicharts/blob/main/data/gpt-subsidy-measurement.json
  6. https://developers.openai.com/api/docs/models/gpt-5.4
  7. https://developers.openai.com/api/docs/models/gpt-5.4-mini
  8. https://developers.openai.com/api/docs/models/gpt-5.6-luna
  9. https://developers.openai.com/api/docs/models/gpt-5.6-terra
  10. https://github.com/junhoyeo/tokscale/blob/0149a44329fb89865837dde40adb8cd9bc06bead/crates/tokscale-core/src/sessions/codex.rs#L98-L214
  11. https://github.com/junhoyeo/tokscale/blob/0149a44329fb89865837dde40adb8cd9bc06bead/crates/tokscale-core/src/sessions/codex.rs#L518-L675