Codex Limit Changes: Official Quota & Policy History
Did Codex limits change? Yes. OpenAI has updated Codex usage limits and tier structures multiple times in 2026, introducing model-dependent consumption dynamics (including GPT-6 Astra and the GPT-5.6 family), flexible pay-as-you-go token-billed seats for teams, removing the 5-hour limit for Business Premium seats, and adding on-demand recovery mechanics like banked and paid instant resets.
If you noticed your Codex limits behaving differently over the last few weeks, three major structural updates took effect:
- •03 Sep 2026 (GPT-6 Astra Integration): Frontier model Astra was added to Codex CLI (v0.153.0+) and desktop apps, drawing directly from your existing Work and Codex allowance pool.
- •24 Aug 2026 (Business Premium Seats): OpenAI removed the rolling 5-hour burst window for Business Premium seats ($100–$125/mo), moving those users to unthrottled weekly resets with 5x usage.
- •Paid Instant Resets (Documented 2026): Eligible personal Plus and Pro subscribers can purchase an immediate weekly rate-limit reset in Usage settings to clear 5-hour and weekly locks.
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, verified against official OpenAI announcements or Help Center documentation, and distinguishes previous behavior from new behavior.
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.
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 specialized GPT-5.6 family to general availability in Codex: GPT-5.6 Sol (flagship complex reasoning, est. 15–90 local messages/5h on Plus), GPT-5.6 Terra (balanced everyday development, est. 20–110 messages/5h), and GPT-5.6 Luna (lightweight high-throughput execution, est. 50–280 messages/5h).
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
Usage-limit resets operated strictly on automated server schedules (rolling 5-hour countdowns and weekly anchors). Users had no mechanism to bank or redeem on-demand promotional resets.
OpenAI rolled out rate-limit reset banking in Settings > Usage for eligible Plus and Pro users, providing one initial complimentary reset grant at launch and the opportunity to earn promotional resets by inviting others to try Codex (with a 30-day expiration window from the grant date). Redeeming a full banked reset instantly restores both 5-hour and weekly Codex usage allowances back to 100% and changes the weekly reset date. Business accounts simultaneously received coworker referral features, distinct from Plus/Pro reset banking.
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.
How We Source and Verify Codex Limit Changes
Unlike community rumor boards or AI aggregators that invent speculative turn figures, Codex Usage strictly categorizes historical entries by source type:
- 1.Official OpenAI Documentation & Announcements: Primary release notes, help center articles (e.g. Paid weekly Work and Codex rate limit resets), and published pricing pages at
chatgpt.com/codex/pricing. - 2.First-Party CLI Telemetry: Concrete parameters extracted directly from authenticated Codex CLI status payloads (
windowDurationMins: 300andwindowDurationMins: 10080). - 3.What We Never Publish: We do not publish speculative weekly turn numbers (e.g., claiming Pro has a fixed 3,000 turn limit), because OpenAI explicitly leaves weekly allowances unpublished.
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 only provides estimated 5-hour message ranges for Plus (varying by model between 15 and 280 local messages) and frames Pro as providing 5x to 20x more usage than Plus. 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 set an OPENAI_API_KEY to switch to metered platform billing.
Looking for currently active tier numbers? Check our live Codex usage limits table by plan or explore the specific breakdown for ChatGPT Plus Codex limits and ChatGPT Business Standard vs Premium seats.
Need to know how recovery windows function? Read when your Codex usage resets or see how promotional resets work in our Codex banked resets guide.
Throttled right now? Compare purchased credits, paid instant resets, and API fallback to resume working immediately.