Skip to content
AI Primer
update

Codex updates paid-plan limits with a weekend reset

Posts said Codex usage limits were reset across paid plans as users shared Mac app feedback, browser control, and repo-review results. The examples show Codex being used as a daily driver for debugging and code audit work, so watch the limits if you rely on it for regular use.

4 min read
Codex updates paid-plan limits with a weekend reset
Codex updates paid-plan limits with a weekend reset

TL;DR

OpenAI also pushed Codex onto phones via steipete's repost of OpenAI, with setup screens in LLMJunky's screenshot and petergyang's setup screenshot showing remote approval, computer-use access, and a Chrome extension hook. The odd little tell in this batch is that the most useful evidence is not a benchmark chart, it's a pile of workflow screenshots: repo audits, limit meters, and mobile review flow.

Weekend reset

The reset posts landed a day apart, which made the contrast hard to miss. ClaudeDevs announced a reset of Claude Code's 5-hour and weekly buckets on May 15, and steipete's repost of Thomas Sottiaux said Codex limits had been reset across paid plans on May 16.

The pressure behind that reset shows up in bas_fijneman's screenshot, where a weekly Codex meter on the $100 plan was already heavily consumed. Several commentary posts in the evidence pool frame subscription access as the unlock, not just model quality, including steipete's repost of danmatern saying use through the subscription changed their main-agent choice.

Mac app workflow

The strongest product feedback came from awilkinson's Mac app review, which compared Codex directly against Claude Code on macOS and argued the visual tab model made parallel work easier to manage.

That same post surfaced five concrete friction points:

  • AskQuestionTool is missing in work mode, so requests for input fall back to long text questions.
  • Activity updates make the app feel slower than it may actually be.
  • Human-style names for sub-agents are distracting.
  • Settings and environment sync across multiple Macs is still awkward.
  • Browser control was unclear enough that awilkinson thought it might be missing.

On that last point, danshipper's reply said Codex can control the integrated browser. That matches the remote-control setup shown later in petergyang's screenshot, which includes toggles for computer use and a Chrome extension.

Repo review

A more interesting signal than the limit drama is what people are actually using the agent for. steipete's clawpatch post pointed to clawpatch.ai and included a repo review that separated one runtime bug from five lower-priority test and maintenance issues.

The findings list in that screenshot breaks down cleanly:

  • A likely ffmpeg pipe deadlock in AttachmentResolver.swift
  • A missing --chat test case
  • Temp SQLite fixture cleanup missing on Linux tests
  • Missing order assertions in helper-resolution tests
  • A weak attachment staging test
  • No smoke test for the actual @main executable

steipete's repost of aleks_todo pushed the same pattern further, claiming a first pass on another repo found 27 issues. That is a much louder pitch than "AI pair programmer." It is closer to automated code audit, with the output delivered as a review report instead of a chat transcript.

Mobile remote control

OpenAI's mobile push turned Codex into a review-and-steer surface away from the desk. According to steipete's repost of OpenAI, the mobile preview lets users start new work, review outputs, steer execution, and stay on top of long-running jobs from the ChatGPT app.

The setup evidence adds a few concrete details that the announcement post only implies:

Further reading

Discussion across the web

Where this story is being discussed, in original context.

On X· 3 threads
TL;DR2 posts
Weekend reset1 post
Mobile remote control3 posts
Share on X