Codex Pricing 2026: Plans, Limits & Banked Reset for Agencies

Published August 23, 2026By ABD Legacy LLC
Codex pricing Codex limits how many people use Codex banked reset

Status: Codex just crossed 20 million active users — and every paid user got a banked reset

Codex and ChatGPT Work crossed 20 million active users during the week of August 21, 2026, per Tibo Sottiaux, OpenAI's Codex & ChatGPT engineering lead — up from 15 million just over a week earlier (Aug 13), and roughly 10 million on July 21.[1][2][5] To celebrate, OpenAI credited every Codex and ChatGPT Work user with a BANKED reset they could use at their leisure.[1]

Rollout status (as of Aug 23, 2026): The reset was promised "during the day" on Aug 21, then given a firmer 8pm PST same-evening deadline for all paid ChatGPT Work and Codex users.[15] 8pm PST came and went with many users reporting no reset; some users did receive it.[2][3] On Aug 23, OpenAI announced it found usage inefficiencies (image-heavy long sessions, Computer History p95+ usage, a title-generation feature) plus a worse cache hit rate, and posted a new reset for 2026-08-24T21:00:00Z (2pm PST).[3]

What a banked reset is (official definition)

Per OpenAI's Help Center, a banked reset is a saved usage-limit reset credit: "Using a full banked reset resets both your 5-hour and weekly usage windows. As a result, your weekly reset date will move to roughly seven days after you redeem it."[4] Banked resets are separate from automatic resets: they appear as available reset credits on the ChatGPT Codex usage page (chatgpt.com/codex/settings/usage) and can be redeemed after you hit a limit, rather than applying immediately.[4][11] Codex is included across ChatGPT plans, including Free and Go; usage limits vary by plan.[4]

Codex pricing in 2026 — the plan table

Verified against official docs via CodePick (2026-08-19). No price changed with the 20M milestone; these are the current list prices.[14]

PlanPriceWhat agencies should know
ChatGPT Plus$20/monthCodex included; 5-hour + weekly usage windows apply.
Pro 5x$100/month5× usage of Plus; heavier agent workloads.
Pro 20x$200/month20× usage of Plus; highest subscription tier for heavy coding.
ChatGPT BusinessPer seatWorkspace seats; usage limits vary per member.
APIUsage-basedPer-token billing; GPT-5.6 Sol promo $4/$20 per 1M through Nov 21, 2026 (down from $5/$30).

Per-model 5-hour window ranges exist (e.g., Sol 10–100 on Plus vs Luna 250–2,000), and credits are token-billed — so the same plan tier can deliver very different per-task economics depending on the model and workload.[14] Re-run your engagement math with current assumptions in the AI Agency Pricing Calculator.

What this means for agency pricing and usage-based billing

For agencies billing AI work, the 20M milestone and the banked-reset saga are a capacity and billing-clarity story, not a price story:

  1. No list price changed. Plus $20, Pro 5x $100, Pro 20x $200, Business per seat, and API usage-based rates all stand. Do not raise client quotes because Codex "got more users" — the milestone did not touch the price sheet.[14]
  2. Limits are the real capacity risk. Codex runs on 5-hour and weekly usage windows that vary by plan, and OpenAI does not publish one fixed weekly allowance that applies to every user and workload.[10] Agencies quoting fixed-fee retainers should verify the allowance before pricing, and plan headcount against standard limits — the Aug 21–23 drain complaints (GitHub issues #36468, #36488) show how fast a heavy agent session can consume a week's quota.[10][1]
  3. Banked resets are a scheduling tool, not a billing line. A banked reset lets a blocked migration or a long agent loop keep running without waiting for the 5-hour window — but it resets windows, it does not change token math. Budget it as capacity insurance, never as a cost line in a client quote.[4][11]
  4. Keep subscription and token lines separate in client quotes. This calculator models API/token and project pricing, not seat-based subscriptions — Codex/ChatGPT Work seats are a budget line of their own, and mixing them into a per-task price hides the real consumption profile.[14]
Watch item: OpenAI has acknowledged usage-limit inefficiencies and worse cache-hit rates this week (Aug 23 update) and is fixing them; the Aug 24 reset announcement includes the exact delivery time for the first time.[3] If limits keep draining abnormally fast after the fix, agencies should re-verify the assumptions behind fixed-fee AI retainers. No deliberate "quiet cut" has been confirmed — treat that as critique, not fact.[2]

Model Codex and ChatGPT Work costs in your next quote

Open the AI Agency Pricing Calculator →

Keep the subscription line separate from token math, and verify limits before pricing fixed-fee retainers.

Frequently asked questions

How many people use Codex?

Codex and ChatGPT Work crossed 20 million active users during the week of August 21, 2026, per OpenAI engineering lead Tibo Sottiaux — up from 15 million just over a week earlier (Aug 13) and 10 million on July 21.[1][2][5] Figures are company-reported, not audited, and most coverage treats 20M as the combined Codex + ChatGPT Work count.[6][18] Full breakdown: Codex Just Hit 20 Million Users — and Every User Got a "Banked Reset".

What is a banked reset in Codex?

A saved usage-limit reset credit. Redeeming a full banked reset resets both your 5-hour and weekly usage windows, and your weekly reset date moves to roughly seven days after you redeem it — per OpenAI's Help Center.[4] It appears in your Codex usage page and is separate from automatic resets.[11]

Did the Codex banked reset actually arrive?

Disputed. OpenAI promised it "during the day" Aug 21, then set a firmer 8pm PST same-evening deadline for all paid ChatGPT Work and Codex users — which passed without the reset appearing for many accounts.[2][15] Some users did receive it.[3] A new reset was announced for 2026-08-24T21:00:00Z (2pm PST) after OpenAI found usage inefficiencies.[3]

What are Codex limits and why do they matter for agencies?

Codex runs on 5-hour and weekly usage windows that vary by plan; OpenAI does not publish one fixed weekly allowance that applies to every user and workload.[10] For agencies, limits are a capacity risk: verify the allowance before quoting fixed-fee retainers, plan headcount against standard limits, and treat banked resets as scheduling insurance, not a billing line.

How much does Codex cost in 2026?

Verified via CodePick against official docs (2026-08-19): ChatGPT Plus $20/month, Pro 5x $100/month, Pro 20x $200/month, ChatGPT Business per seat, and API usage-based billing.[14] OpenAI's published GPT-5.6 Sol API promo rate is $4 per 1M input / $20 per 1M output through Nov 21, 2026 (down from $5/$30).

Should the 20M-user milestone change what agencies charge?

No list price changed with the milestone, so pricing math stands. What changes is capacity and billing clarity: 20M users means more limit pressure, banked resets become part of the usage story, and agencies should keep token and seat lines separate in client quotes rather than folding subscription access into a fixed per-task price.[14]

Sources

Accuracy note: 20M active users = company-reported by Tibo Sottiaux, not audited; most coverage treats it as combined Codex + ChatGPT Work (flagged). "8pm PST missed" reflects user reports via explainx and community threads — some users received the credit. Aug 24 21:00Z reset announced Aug 23 and needs re-verification after it lands. The $4/$20 Sol API rate is OpenAI's published promo (Aug 21 – Nov 21, 2026). "Quietly cut rate limits" is Ed Zitron's characterization, quoted secondhand, not an OpenAI admission. Plan prices verified via CodePick 2026-08-19; re-verify before quoting.