Codex Limit Changes: Quota, Reset & Policy History
Did OpenAI Codex limits change? Yes. OpenAI changes Codex limits, quota rules, pricing, and reset mechanics on an irregular schedule. In 2026, OpenAI introduced model-specific usage estimates, removed the 5-hour window for Business Premium seats, cut metered token rates by up to 80%, launched banked and paid instant resets, and applied global allowance resets during the GPT-6 Astra rollout. This ledger keeps a dated record of material confirmed changes, official documentation updates, and corroborated observed incidents.
If your Codex limits or reset counters behaved differently over the last two weeks, several major events took effect:
- •09–11 Sep 2026 (Observed anomalies): Developers across multiple accounts reported unexpected quota changes, shifting reset timestamps, and conflicting limit states in the official openai/codex repository, with related reports continuing through September 11. OpenAI has not confirmed a common cause.
- •10 Sep 2026 (Pro $200 pause): OpenAI temporarily paused new sign-ups and upgrades to ChatGPT Pro $200 (Pro 20X); existing subscriptions continue renewing normally.
- •09 Sep 2026 (Deep Research in Work & Codex): Deep Research became available in ChatGPT Work and Codex, drawing from existing Work and Codex allowances or credits.
- •07 Sep 2026 (Global subscription reset): OpenAI applied an automatic one-time quota reset for eligible Plus, Pro, and Business users during the GPT-6 Astra rollout.
- •03–04 Sep 2026 (Banked reset grants): OpenAI granted banked resets to eligible Plus, Pro, and Business accounts on both September 3 and September 4.
- •03 Sep 2026 (GPT-6 Astra integration): Astra launched across Work and Codex surfaces and draws from existing Work and Codex allowance rather than an isolated Astra quota.
Why Codex Limits Feel Different: Model Dynamics & Usage Factors
The most common reason developers ask “did OpenAI reduce my Codex limits?” is not a published quota reduction, but how modern model selection, reasoning intensity, and task size impact allowance consumption compared to simple text queries.
OpenAI documents that consumption against your included allowance varies with model choice, reasoning effort, task complexity, context length, and input/output size. While OpenAI does not publish an exact quota formula for Plus/Pro allowances, frontier reasoning sessions draw down rolling windows much faster than lightweight tasks.
Explicit token-based metering applies to flexible pay-as-you-go seats (introduced for Business/Enterprise teams) and credit top-ups. These track uncached input tokens, cached context, and generated output tokens against published rate cards rather than drawing from subscription allowances.
Chronological Codex Limit Change Ledger
Every entry in this ledger is dated and assigned an evidence status. Confirmed changes and documentation updates are backed directly by dated OpenAI releases, Help Center articles, or official announcements. Observed incidents require corroborating first-party telemetry or multiple concrete issue reports and remain explicitly unconfirmed until OpenAI documents a root cause or policy change.
Directly verified by dated OpenAI documentation or an official announcement.
Help Center revision or clarification where behavior may predate publication.
Supported by multiple concrete developer reports or telemetry; unconfirmed by OpenAI.
Quota and Reset-State Anomalies Reported Across Multiple Codex Accounts
Users rely on Settings > Usage, /status, and server-reported reset timestamps to understand the current state of their 5-hour and weekly Codex allowances. From September 9 through September 11, multiple reports documented abrupt inconsistencies between those quota readings, reset timestamps, and subsequent request behavior.
Multiple developer reports submitted to the official openai/codex GitHub repository documented related quota and reset-state anomalies between September 9 and September 11, 2026. Reported behavior included weekly remaining quota changing sharply while an account was idle, weekly reset timestamps moving by several days, quota states changing and later reverting, requests being blocked while other usage indicators appeared inconsistent, and banked-reset state being affected around later quota recalculation.
Related reports continued through September 11, adding further independent observations of unexpected quota depletion and inconsistent quota state—including allowance falling while idle or immediately after relaunching Codex across separate Plus and Pro accounts, alongside references to related reset mismatches and suspected overcounting.
These reports describe related symptoms across multiple independent accounts, but they do not establish a single incident, one server bug, an accounting defect, a new reset algorithm, or an unannounced policy change. While backend reconciliation or accounting issues were raised as possible explanations by reporters, OpenAI has not publicly confirmed a common root cause, double counting, or a policy modification.
ChatGPT Pro $200 New Sign-Ups and Upgrades Temporarily Paused
Users on Free, Go, Plus, and Pro $100 plans could purchase or upgrade directly to the ChatGPT Pro $200/month (Pro 20X) subscription tier.
OpenAI updated its ChatGPT Pro availability policy to temporarily pause new sign-ups and upgrades to ChatGPT Pro $200 (Pro 20X) as of September 10, 2026. The pause applies to moves into Pro $200 from Free, Go, Plus, and Pro $100.
Existing Pro $200 subscriptions are unaffected and continue renewing normally, with existing usage allowances unchanged. Pro $100 remains available for new purchases and upgrades. If an existing Pro $200 subscriber cancels or completes a downgrade and the Pro $200 subscription ends, they cannot purchase Pro $200 again until the pause is lifted.
Deep Research Integrated into ChatGPT Work and Codex Allowance Pool
Deep Research was not available within ChatGPT Work and Codex environments.
OpenAI expanded ChatGPT Work and Codex to support Deep Research workflows. In its September 9, 2026 release notes, OpenAI confirmed that Deep Research became available in ChatGPT Work and Codex and consumes the user's existing included Work and Codex allowance or available credits, rather than operating under a separate quota pool.
Global Codex Usage Reset for Paid Subscriptions
Users who had exhausted or heavily consumed their current Codex usage allowance would normally remain constrained until their applicable allowance recovered, reset, or another eligible recovery mechanism was used.
OpenAI announced a global usage reset for paid subscriptions, expected around 6 PM Pacific on September 7, restoring eligible usage directly so users could continue using Astra. This was an automatic global reset applied directly to eligible usage limits and did not create a saved reset for later use—unlike a promotional banked reset, which is saved to the account and must be manually applied by the user.
Banked Codex Resets Granted to Eligible Accounts on September 3 and September 4
Banked resets already existed as a promotional recovery mechanism. During the GPT-6 Astra rollout, OpenAI added two new dated banked-reset grants with specific eligibility cutoffs.
OpenAI granted promotional banked resets across two consecutive GPT-6 Astra rollout waves.
On September 3, eligible existing Plus, Pro, and Business users in good standing received a banked reset. New Plus, Pro, and Business accounts, including qualifying Standard and Premium seats, were eligible if created before 10 PM PT that day.
On September 4, eligible existing users of the same plans received another banked reset. New qualifying accounts and Business seats were eligible if created before 8 PM PT that day.
A banked reset is saved to the account for manual use. Applying a full banked reset refreshes the 5-hour and weekly Codex usage windows and changes the weekly reset date. OpenAI now clarifies that the saved reset is consumed only when it successfully refreshes at least one eligible usage window. Expiration depends on the specific offer and is shown in the account.
These saved resets were distinct from the September 7 global reset, which OpenAI applied directly to eligible usage limits and which could not be banked for later.
GPT-6 Astra Integrated into Unified Work & Codex Allowance Pool
Codex sessions ran on GPT-5.6 family checkpoints (Sol, Terra, Luna) and GPT-5.5. No GPT-6 reasoning model was accessible within Codex CLI or desktop agent environments.
Frontier reasoning model GPT-6 Astra integrated across Codex surfaces. Astra does not receive an independent message cap; instead, it draws directly from existing Work and Codex allowance pools. Plus accounts include limited Astra access; Pro ($100 & $200) accounts can use their full existing allowance for Astra. API usage is metered per token ($10/M input, $50/M output standard; $20/M input, $100/M output Fast mode).
ChatGPT Business Premium Seats Launched: 5-Hour Limit Removed
All ChatGPT Business workspace members held Standard seats ($20–$25/seat/month), subject to rolling 5-hour burst windows and standard weekly reset cycles.
OpenAI introduced Premium seats ($100/seat/month billed annually or $125 billed monthly). Premium seats eliminate the standard rolling 5-hour limit entirely, transition to predictable weekly usage resets, provide 5x more included usage than Standard seats, and include full access to GPT-6 Astra. Teams can mix Standard and Premium seats in the same workspace.
GPT-5.6 Sol API and Credit Pricing Reduced by Over 20%
At GPT-5.6 launch, Sol was priced at $5 per million input tokens and $30 per million output tokens.
OpenAI reduced GPT-5.6 Sol API and credit pricing by more than 20% for the following three months. Sol moved from $5 to $4 per million input tokens and from $30 to $20 per million output tokens.
The change lowered the metered cost of using Sol through the API and eligible ChatGPT Work and Codex credit usage. Included Plus, Pro, and Business subscription usage remained unchanged.
GPT-5.6 Luna and Terra Metered Pricing Reduced by 80% and 20%
At GPT-5.6 launch, Terra was priced at $2.50 per million input tokens and $15 per million output tokens. Luna was priced at $1 per million input tokens and $6 per million output tokens.
OpenAI cut GPT-5.6 Luna pricing by 80%, to $0.20 per million input tokens and $1.20 per million output tokens, and cut GPT-5.6 Terra pricing by 20%, to $2 per million input tokens and $12 per million output tokens.
The lower prices also reduced how much eligible Luna and Terra usage consumes from paid ChatGPT Work and Codex credits. ChatGPT and Codex subscription prices and quota budgets remained unchanged.
The same announcement introduced GPT-5.6 Sol Fast mode in the API, offering up to 2.5 times faster performance at twice the standard token price.
Paid Weekly Rate-Limit Resets Documented for Personal Plus & Pro
Standard rate-limit recovery relied on automated rolling 5-hour windows, scheduled weekly resets, promotional banked resets when available, or custom API key billing.
Eligible personal Plus and Pro subscribers can purchase an on-demand paid instant reset via desktop Usage settings or weekly-limit banners. Purchasing pulls the next weekly allowance forward immediately, clears both 5-hour and weekly locks, and begins a new 7-day weekly reset period starting from the first Work or Codex request sent after the reset is applied. Business, Enterprise, Edu, and Free tiers are ineligible.
GPT-5.6 Model Family (Sol, Terra, Luna) GA in Codex with Tiered 5-Hour Estimates
Codex default execution relied on prior generation checkpoints (such as GPT-5.5 / GPT-5.4) prior to the specialized GPT-5.6 tier architecture.
OpenAI released the GPT-5.6 model family to general availability in Codex, introducing Sol, Terra, and Luna with model-specific usage economics. At launch, OpenAI's Codex pricing guidance published the following estimated 5-hour local-message ranges for ChatGPT Plus:
GPT-5.6 Sol: 15 to 90 local messages per 5 hours.
GPT-5.6 Terra: 20 to 110 local messages per 5 hours.
GPT-5.6 Luna: 50 to 280 local messages per 5 hours.
These are historical launch-era estimates, not today's active numbers. OpenAI later changed its published guidance without providing a dated effective-change notice in the current documentation. Current Plus estimates are 10 to 100 for Sol, 25 to 200 for Terra, 250 to 2,000 for Luna, and 5 to 45 for Astra. See the current Codex usage limits table, or the GPT-5.6 Luna limits page, for up-to-date ranges.
Standalone Legacy Codex Seats Grandfathered for Business Workspaces
Any ChatGPT Business workspace could add standalone, usage-based Codex seats without a monthly seat fee, drawing activity exclusively from shared workspace credits.
Standalone usage-based Codex seats were discontinued for new workspaces. Workspaces created after June 24, 2026, or workspaces that had never added a Codex seat prior to that date, were required to assign Standard seats to use Codex (workspaces today may assign Standard or Premium seats). Workspaces with active Codex seats or pending invites before June 24 retain grandfathered management rights.
Rate-Limit Reset Banking Launched for Eligible Plus & Pro Accounts
Before reset banking, users relied on normal 5-hour and weekly allowance recovery or other available billing paths. There was no saved promotional reset that could be held in an account and manually applied later.
OpenAI introduced rate-limit reset banking for eligible Plus and Pro users on June 11, including one free reset at launch and a referral flow that could award both the referrer and an eligible recipient a banked rate-limit reset after the recipient sent their first Codex message. The release notes stated that these promotional resets were usable for 30 days after being granted.
At launch, OpenAI described a full banked reset as refreshing the user's Codex usage allowance. Current Help Center documentation clarifies the mechanic further: a full banked reset refreshes the 5-hour and weekly Codex usage windows, changes the weekly reset date, and is consumed only when it successfully refreshes at least one eligible window. If there is nothing eligible to reset, the saved reset remains available.
See the complete Codex banked resets guide for current redemption rules.
Codex-Only Pay-As-You-Go Seats & Token Billing Introduced for Teams
Business and Enterprise workspace members accessed Codex through standard subscription seats with included plan usage limits. Standalone seats billed purely on token consumption were not available.
OpenAI introduced flexible Codex-only pay-as-you-go seats for Business and Enterprise workspaces, allowing teams to add seats without a monthly per-user fee and pay according to actual token consumption (input tokens, cached tokens, and generated output tokens). Standard Business seats continued to receive included Codex usage limits, and certain Enterprise and Edu workspaces remained on legacy metering during migration.
GPT-5.3-Codex-Spark Research Preview Launched with Dedicated Rate Limits
Codex sessions drew exclusively from standard subscription allowances and unified 5-hour rolling burst and weekly reset windows across available models.
OpenAI introduced GPT-5.3-Codex-Spark as a research preview for ChatGPT Pro users. Spark operates under its own separate rate limits during the research preview and usage does not count against standard Codex rate limits. OpenAI notes that separate rate limits may change based on demand, and access may be temporarily queued during high traffic periods.
Where these entries come from
Every ledger entry is categorized by the evidence supporting it.
- 1.Official OpenAI documentation and announcements: Product releases, Help Center articles, release notes, and published OpenAI pricing or rate-card documentation.
- 2.First-party telemetry: Concrete runtime values exposed by OpenAI clients or server responses, such as usage percentages, reset timestamps, and documented rate-limit-window metadata.
- 3.Observed incidents: Multiple concrete reports in first-party-adjacent surfaces such as the official
openai/codexGitHub issue tracker. These entries remain explicitly unconfirmed until OpenAI documents a root cause or policy change. A filed GitHub issue proves that behavior was reported, not that OpenAI has confirmed its cause. - 4.Speculative numbers: None. This page does not invent unpublished weekly quotas or present Reddit posts, generic AI blogs, or secondary roundups as primary evidence.
Frequently Asked Questions About Codex Limit Changes
Direct answers to common questions about recent limit adjustments and quota tracking.
Did OpenAI reduce Codex usage limits recently?
OpenAI has not published an official reduction in base plan allowances. However, usage consumption against included plan allowances varies significantly based on model choice, reasoning effort, and task complexity. Frontier reasoning models (like GPT-5.6 Sol or GPT-6 Astra) and prompts with deep repository context consume 5-hour and weekly allowances substantially faster than lightweight tasks. For pay-as-you-go seats and credit top-ups, usage is metered directly on token consumption.
Why am I hitting my 5-hour Codex limit faster than before?
OpenAI documents that usage consumption depends on several factors—including model selection, task complexity, reasoning effort, context length, and input/output size—rather than raw prompt counts. Running complex agentic refactors or reasoning-heavy frontier models draws down your 5-hour window far faster than short questions. (Direct per-token metering applies when utilizing pay-as-you-go seats or credits).
Does OpenAI publish exact weekly turn caps for Plus or Pro?
No. OpenAI does not publish exact weekly token or turn caps for any tier. Official documentation provides model-specific estimated 5-hour message ranges for Plus, currently spanning 5 to 2,000 local messages depending on the model, and states that Pro provides either 5x or 20x more usage than Plus depending on the Pro tier. Your live percentage and reset anchor are visible via /status.
How do I check if my Codex reset time or limits have changed?
Run /status inside the Codex CLI, or navigate to Settings > Usage in the Codex desktop app or ChatGPT web settings. The CLI telemetry reports your exact primary (5-hour) and secondary (weekly) remaining percentages alongside live Unix reset timestamps directly from OpenAI servers.
What should I do if I hit my limit before my weekly reset?
If you hit the 5-hour burst limit, wait for rolling recovery. If you hit the weekly cap, you can redeem an available promotional banked reset (Settings > Usage), purchase a paid instant reset (eligible personal Plus/Pro accounts), draw from purchased ChatGPT credits, or authenticate the Codex CLI with an OpenAI API key to switch to metered Platform billing.
What does an "Observed incident" entry on this page mean?
An Observed incident means multiple concrete developer reports or first-party telemetry describe related unusual quota or reset behavior, but OpenAI has not publicly confirmed a common root cause or policy change. These entries are visually separated from confirmed changes and documentation updates and are never presented as established policy.
Did OpenAI change how or when the Codex weekly limit resets?
OpenAI has documented several mechanisms that can change a weekly reset date. A full banked reset refreshes the eligible 5-hour and weekly windows and changes the weekly reset date. Eligible personal Plus and Pro accounts can also purchase an instant reset, which starts a new weekly period on the first Work or Codex request after the reset is applied, with the next automatic reset seven days later. Business Premium seats use weekly resets without the Standard seat's 5-hour limit. Outside those documented mechanisms, the reset-timestamp changes reported between September 9 and 11 are recorded here as an Observed incident, not an announced policy change.
- Codex usage limits table by plan — the numbers in force today.
- ChatGPT Plus Codex limits — the Plus breakdown.
- ChatGPT Business Standard vs Premium seats — seat-level differences.
- when your Codex usage resets — how recovery windows work.
- Codex banked resets guide — promotional grants.
- purchased credits — pay-as-you-go overflow.
- paid instant resets — buying your next week early.