Currentness: Source audit is overdue or does not match this snapshot. This is a modeled decision aid, not a current recommendation.
Plus/Pro is appropriate for interactive Codex work that fits shared rolling ChatGPT plan headroom. API/BYO is safer for bursty, CI, team, programmatic, or spend-controlled workflows.
Local Codex CLI and ChatGPT cloud Codex share ChatGPT plan headroom. OpenAI publishes five-hour estimate ranges and states weekly rolling caps exist, but the exact weekly ceiling is not published here as a number.
Codex pricingUsing Codex with ChatGPT planOpenAI API pricing
Decision: Trial Codex for the work it appears to fit, but verify quota telemetry and Desktop stability on your model, plan, and operating system before switching.
Interpret Plus and Pro as subscription headroom for local + cloud Codex, not as a guaranteed API-style token allotment.
Codex CLI and ChatGPT cloud Codex draw from ChatGPT plan headroom.
Using Codex with ChatGPT planA weekly rolling cap exists; no exact weekly number is shown here.
Codex pricingMetered API usage is a spend-control decision, not a ChatGPT cap.
OpenAI API pricingGPT-5.5 · ~15–80 GPT-5.5 msgs / 5h (local+cloud shared)
estimateshared local + cloudnot a guarantee
OpenAI publishes five-hour estimate ranges and a weekly rolling-cap concept, not an exact weekly ceiling to copy here.
GPT-5.5 · ~75–400 GPT-5.5 msgs / 5h
estimateshared local + cloudnot a guarantee
OpenAI publishes five-hour estimate ranges and a weekly rolling-cap concept, not an exact weekly ceiling to copy here.
GPT-5.5 · ~300–1600 GPT-5.5 msgs / 5h
estimateshared local + cloudnot a guarantee
OpenAI publishes five-hour estimate ranges and a weekly rolling-cap concept, not an exact weekly ceiling to copy here.
Range → selected point: OpenAI publishes ~15–80 GPT-5.5 messages / 5h. The model selects 45 messages, an editorial central estimate inside the range—not the low bound, high bound, or exact arithmetic midpoint. OpenAI Codex range and window · 2026-06-24 ↗
Monthly utilization assumption: 40 active hours/week ÷ 5h × 52/12 = 34.67 five-hour windows/month. The 25,000 tokens/message input is an adjustable editorial workload assumption for context, tool results, history, and output—not a measured average. Change the workload assumption.
Tokens/month: 45 × 25,000 × 34.67 = 39.00M (displayed 39M). Effective $/M: $20 ÷ 39.00M = $0.513/M.
API baseline: GPT-5.5 direct API at 15:1 input/output: ($5.00 × 15 + $30.00) ÷ 16 = $6.5625/M. OpenAI GPT-5.5 API rates · 2026-06-24 ↗
API equivalent: $6.5625/M ÷ $0.5128/M = 12.80× (displayed 13×).
Range → selected point: OpenAI publishes ~75–400 GPT-5.5 messages / 5h. The model selects 230 messages, an editorial central estimate inside the range—not the low bound, high bound, or exact arithmetic midpoint. OpenAI Codex range and window · 2026-06-24 ↗
Monthly utilization assumption: 40 active hours/week ÷ 5h × 52/12 = 34.67 five-hour windows/month. The 25,000 tokens/message input is an adjustable editorial workload assumption for context, tool results, history, and output—not a measured average. Change the workload assumption.
Tokens/month: 230 × 25,000 × 34.67 = 199.33M (displayed 199M). Effective $/M: $100 ÷ 199.33M = $0.502/M.
API baseline: GPT-5.5 direct API at 15:1 input/output: ($5.00 × 15 + $30.00) ÷ 16 = $6.5625/M. OpenAI GPT-5.5 API rates · 2026-06-24 ↗
API equivalent: $6.5625/M ÷ $0.5017/M = 13.08× (displayed 13×).
Range → selected point: OpenAI publishes ~300–1600 GPT-5.5 messages / 5h. The model selects 900 messages, an editorial central estimate inside the range—not the low bound, high bound, or exact arithmetic midpoint. OpenAI Codex range and window · 2026-06-24 ↗
Monthly utilization assumption: 40 active hours/week ÷ 5h × 52/12 = 34.67 five-hour windows/month. The 25,000 tokens/message input is an adjustable editorial workload assumption for context, tool results, history, and output—not a measured average. Change the workload assumption.
Tokens/month: 900 × 25,000 × 34.67 = 780.00M (displayed 780M). Effective $/M: $200 ÷ 780.00M = $0.256/M.
API baseline: GPT-5.5 direct API at 15:1 input/output: ($5.00 × 15 + $30.00) ÷ 16 = $6.5625/M. OpenAI GPT-5.5 API rates · 2026-06-24 ↗
API equivalent: $6.5625/M ÷ $0.2564/M = 25.59× (displayed 26×).
Choose ChatGPT Plus/Pro Codex when work is interactive. The upside is predictable monthly spend; the downside is rolling-limit drain across the shared local + cloud Codex bucket.
Choose API/BYO when work is bursty, CI-driven, team-distributed, programmatic, or needs spend attribution. ChatGPT subscription rolling caps are not the API risk.
| Date | Source | Claim used | Label | Link |
|---|---|---|---|---|
| 2026-06-24 | OpenAI Codex pricing | ChatGPT Plus is $20/mo with an estimated ~15–80 GPT-5.5 messages per 5 hours; ChatGPT Pro 5× is $100/mo with ~75–400; ChatGPT Pro 20× is $200/mo with ~300–1600. | official price + range | open source |
| 2026-06-24 | OpenAI ChatGPT plan help | Local Codex CLI and ChatGPT cloud Codex usage share ChatGPT plan headroom; treat the five-hour window as one shared local + cloud bucket. | official plan mechanics | open source |
| 2026-06-24 | OpenAI Codex pricing | Weekly rolling caps exist for Codex subscription usage, but the exact weekly ceiling is not published; this page does not invent one. | unpublished cap | open source |
| 2026-06-24 | OpenAI API pricing | OpenAI API/BYO usage is metered separately by API pricing, making spend control the risk model instead of a ChatGPT subscription quota wall. | metered API escape | open source |