Codex users report 20% of a $200 paid plan used in one day
Several Codex subscribers say their paid allowances are falling rapidly during ordinary coding work. One user reported consuming 20% of a $200 plan in a day, while another reported 40% spent on a separate project.
TL;DR
- Paid Codex allowance is reportedly draining within a day: AIandDesign's one-day report puts one $200 plan at 20% gone, while LuoErik8lrl's quota report recalls a weekly allocation spent on its reset day.
- A separate weekend project consumed 40% of a $200 plan without producing a result, AIandDesign's weekend report says.
- OpenAI's Codex plan documentation says Codex, Work, Excel, and Workspace Agents share an allowance and credit pool; AIandDesign's plan-limits screenshot labels those same four surfaces as shared.
- Some accounts are shifting implementation and delegation to Muse: AIandDesign's delegation post says a fresh Codex quota would last longer that way, while AIandDesign's Muse post describes a tricky build going well on Muse.
- A reset alert may not mean a fresh allowance: AIandDesign's reset notification led to a screen where AIandDesign found only the existing reset.
An OpenAI developer-forum report logged a ChatGPT Work workflow consuming 3-4x more allowance overnight, and said its dashboard attributed 98% of agentic usage to Work rather than Codex. Dockset's AI Usage manual describes a Mac dock widget that reduces Codex, Claude, and Grok to one remaining-allowance tile, then opens individual reset windows on click.
The 20% and 40% reports
AIandDesign described 20% of a $200 plan disappearing in one day despite doing most coding with Muse. A later post put a separate, ultimately unsuccessful weekend project at 40%.
Neither post includes token counts, task logs, or a model-by-model breakdown. AIandDesign's follow-up says Astra felt unusually generous for its first five days, then reached 67% usage in under 48 hours; in another reply, the same user called a week on the $200 tier impossible even with light use.
After moving from Claude, AIandDesign said they might move back if the experience continued.
Four products, one allowance
OpenAI's plan documentation says Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents share both an allowance and a credit pool. A remaining weekly percentage therefore does not identify which one of those surfaces consumed it.
The developer-forum report adds a useful split: its author said 98% of their agentic usage on September 18 was Work and 2% was Codex.
Dockset AI Usage
In benjaminakar's Dockset demo, circular rings put Codex, Claude, and Grok allowances directly in a custom macOS dock.
The AI Usage manual says its main tile shows the lowest remaining allowance across a provider's main windows; a click reveals each window and its reset countdown. Codex model limits appear in those details.
A reset alert that did not add a reset
The ChatGPT app showed AIandDesign a notification saying a new rate-limit reset was available.
After opening the reset list, AIandDesign said it contained only the reset already on the account.
Astra, Sol, and token accounting
Sol did not appear to solve the depletion for AIandDesign, who said in a follow-up that it also made the $200 quota fall quickly. LuoErik8lrl's thread asked whether Astra had become more expensive or whether the total allowance had changed, but supplied no accounting data to choose between them.
The developer-forum report describes an overnight 3-4x jump in allowance use for the same Work workflow. OpenAI's Codex rate card says flexible Codex pricing moved from per-message billing to API-token usage on April 2, 2026.
Muse delegation
Users in the thread described a split workflow rather than a full model switch:
- AIandDesign's Meta Muse post says implementation was being redirected to Meta Muse 1.3.
- A replenished Codex quota would last longer through delegation to Muse, AIandDesign said.
- Hermes entered the discussion when AIandDesign asked whether it was being used to delegate work.
- On a difficult project, AIandDesign said Muse was doing a good job.
20x access
LLMJunky reported that some accounts had again been allowed to upgrade to the $200, 20x tier, while saying that the option was unavailable on their own account.
The attached plan screen shows a selected $100 5x tier with 20x beside it, making the access question part of the same user frustration as the rapidly falling allowance.
120,000 particles at 60 FPS
LuoErik8lrl's build post says their artificial-life simulation runs 120,000 particles at 60 FPS on an RTX 4090, built in a “100% vibe coded” workflow while exploring emergent design.