Claude account bans and restrictions on benefits, OpenAI takes the opportunity to securely catch you with Codex

By APPSO

The whole world has long suffered from Company A.

This is after Anthropic has been continuously rolling out various features in recent times, but at the same time keeps tightening usage restrictions—readers’ most common reaction in the comment section.

In the “Big Three” (OpenAI, Google, Anthropic), it’s already the one with the strictest usage limits on the other side; and now it’s adding more identity verification—real-name registration is required to use it. Early this morning, they also cut off Pro ($20/month) users’ access to Claude Code.

Anthropic’s head of growth came out to respond, saying they are conducting small-scale tests on about 2% of new professional user registrants, with existing Pro and Max users unaffected; and stating that their current subscription plans cannot cope with users’ massive Token consumption, as they are researching new paid options.

Source:

OpenAI on its side also immediately responded to the controversy over Claude Code kicking out Pro members. A Codex executive, Rohan Varma, directly went head-to-head with Claude Code competitors, and even used the same posting format as Claude Code.

Source:

Anthropic is testing more expensive plans for 2% of users, while Codex is testing with 100% of users—allowing both free and paid tiers to use Codex. And it even added a mischievous line: “Claude Code users are unaffected.”

Claude Code users PAY (pay), Codex users PLAY (play)

Another Codex executive, Tibo, also posted on X saying Codex will continue offering a free version and a PLUS version ($20/month). He also mentioned that OpenAI has enough compute power and powerful models to support Codex’s operation.

Altman also retweeted this post, saying, “We hope you can have plenty of AI.”

Source:

Codex’s reputation on social media has generally not been too bad, especially after OpenAI’s recent “big money giveaway.” First, it reset usage limits across all subscription plans, so everyone could experience the relevant plugins Codex rolled out.

In early April, Codex found that the frequency of users reaching usage limits increased, and they couldn’t find the underlying reason—so they simply reset all users’ usage quota limits. A few days ago, to celebrate Codex’s anniversary and the launch of new features, they reset usage limits for all plans again.

Today, Codex’s executive and Altman posted again, saying that in less than two weeks Codex gained 1 million new users. To celebrate it, Codex’s rate limits were reset again—yet again.

Source:

As early as the day last week when Anthropic released Opus 4.7, Codex had already updated a large batch of important features: Computer Use, a built-in browser, persistent memory, and more than 90 plugins.

These updates are almost directly targeted at Claude Cowork’s features, transforming Codex from a tool that sounds like it’s meant for developers into an efficiency-assistant tool that’s applicable to all scenarios on a computer.

Yesterday, based on the memory feature it previously launched, Codex also rolled out a research preview feature called “Chronicle,” enabling AI to read our screens and organize what we’ve been doing recently into memories.

Codex no longer relies only on chat history to understand context. Combined with the recent screen content it reads, when we send “this” and “that,” Codex can know exactly what we mean.

The newly released GPT Image 2 today has also been integrated into Codex. We can generate and iterate on images within Codex. In a single workflow—from product prototypes, front-end design, to visual concept art and game development tasks—we can use GPT Image 2 to quickly generate visual elements.

If your Claude account is always getting banned and you can’t use the official Claude Cowork or the Claude Code desktop version, or if you’re among that 2% of new users who still can’t access Claude Code even after opening a Pro ($20/month) membership, you may want to try Codex made by OpenAI.

From code tool to all-purpose assistant

What matters most among Codex’s recent updates is the Computer Use it released last week. This capability isn’t exactly new; previously, the model itself had Computer Use capability. Now it requires the tool side to have the corresponding support as well, so the model’s capability can be fully leveraged.

Essentially, it’s an Agent tool that can operate a computer the way humans do—through visual recognition, clicking, and input—so it can autonomously control various applications on the computer.

Previously, when Codex operated software on a computer, it did so by executing different app tasks through certain commands. Overall, it was more like us calling out “Siri, what’s the weather like tomorrow?” to do relatively simple tasks.

With the capability of Computer Use, it not only supports tools that invoke APIs or terminal commands, but can also actually help us complete real actions on our computers—especially suitable for front-end debugging, application testing, and software that doesn’t have open APIs for operation.

And it supports multiple agents working in parallel on a Mac without affecting our normal use of other applications.

It’s worth noting that the Computer Use capability only supports macOS 15 and above. When we tested Codex on our computer (macOS 14.6.1), an error report for SkyComputerUseClient automatically popped up.

In addition, Codex now supports a built-in browser, allowing it to better handle Web scenarios. The web pages generated in Codex can be annotated directly on the page, giving Codex more precise instructions for actions. This is very useful for rapid iteration in front-end, application, and game development.

From Coding, design, lifestyle, productivity to research, Codex now has a rich plugin system to handle various tasks

This update also adds more than 90 plugins and richer tool integrations, enabling Codex to connect to more tools, get more context, and execute actions across platforms. The mentioned popular plugins include Atlassian Rovo (JIRA), the Microsoft suite, Neon by Databricks, Remotion, Render, Superpowers, and so on.

In the Codex app, we only need to type a slash to quickly access some Codex configuration options. If we type $, we can choose different Skills, including various Skills we install locally.

At the same time, for automation tasks: after upgrading Codex’s Automation feature, it can reuse previous conversation threads, retaining existing context. The new automation also supports Codex autonomously planning subsequent work, automatically continuing tasks at some future time, and supporting long-term tasks that can last for days or even weeks.

The official says this update is mainly for tasks like code submission merging, following up on everyday work and life to-dos, and information tracking across different platforms and tools.

There are also some small updates for desktop app interactions, such as adding a terminal window with multiple tabs, a sidebar that can directly open files, and previews for documents like PDFs, spreadsheets, and PPTs.

A new summary panel can also continuously track the current task plan and progress, reference information sources, and output results. These enhancements on the app side make Codex more like a unified workspace overall, rather than just a single chat window.

Maintain Agent memory with scheduled screenshots

Personalized memory has long been a major challenge for AI. Although AI can remember knowledge from every era, for each user’s private memory, working memory, and so on, AI needs a way that doesn’t consume too many Tokens while still being able to remember day after day in a way that stays clear.

Especially for tasks like Agents that consume huge numbers of Tokens, the context each user generates every day—if the Agent is to remember all of it—then even a million Tokens’ worth of context wouldn’t be enough to handle.

Last week, OpenAI already brought a memory feature to Codex. It can remember our personal preferences, previous edits we made, and some important information that’s not easy to obtain.

And to get more memory and process our workflows faster, Codex this time launched the Chronicle feature. Put simply, it means looking at our screens, remembering our work, and then feeding these memories back to the AI.

Specifically, in Codex settings > Personalization, once Chronicle is enabled, it will automatically carry out these operations: screen context capture → local temporary screenshot → background agent analysis → temporary Codex session summary → generate local Markdown memory → use it as context in subsequent sessions.

After Codex gets screen recording and accessibility permissions, Chronicle will run a sandbox Agent in the background. These Agents use the default model GPT-5.4-mini. Based on the captured screen images, they periodically start a temporary Codex session, organizing the most recent screen context into memory.

Screen screenshots are only saved temporarily on the local machine. Codex mentions that during operation, any screenshots older than 6 hours will be automatically deleted.

GPT Image 2 generated infographics

In the future, when we talk with Codex, it will automatically retrieve these memory files and use them as context, reducing the need for us to repeatedly describe background details.

OpenAI’s official also provided multiple examples. For instance, if Chronicle isn’t enabled, Codex doesn’t know what our statement “this will fail here” refers to.

And for some personal tasks involving names, project names, and other content beyond general knowledge, Codex will also automatically supplement context based on the information obtained via Chronicle.

Being able to capture screen images also means that the entire workflow of using Codex to handle tasks can be remembered. This includes our workflows and commonly used tools. In the example below, a Codex using Chronicle will know what format that promotional material uses—and which tool it’s from, whether it’s Google Docs or a Markdown document.

However, this feature also faces some controversy. For example, the method of visual recognition consumes a large number of tokens, and more seriously, these screenshots may contain sensitive information visible on our screens.

Although OpenAI says that all saved memories are stored in local markdwon documents, users can view them at any time, and Codex can know what information it obtained from those screenshots— they also remind users that when Chronicle takes screenshots of some risky websites, the websites may hide malicious instructions on the screen through prompt injection, causing Codex to execute them.

At present, Chronicle is only available to ChatGPT Pro ($200/month) users. It supports the macOS version of the Codex app and is released as a research preview. After Chronicle officially goes live, it’s expected that Codex will open it up to more users.

Mobile remote control, electronic pets, and the “Hermes Agent” could all roll out

Recently, Codex has been called by netizens as a product that’s really trying hard to catch up to Claude. On one hand, it’s saying that OpenAI has no direction and just follows the crowd. But on the other hand, seeing fierce competition where good products chase after one another is also not necessarily a bad thing for users like us.

Codex developers asked on X what everyone thinks about Codex, and netizens were very enthusiastic, saying it should add a mobile control feature. Others said Codex should also enter through the ChatGPT App—features that Claude has already achieved.

Some netizens also responded below that Codex has various bugs, such as memory leaks and sessions that can only be archived and not deleted.

In the latest Codex update leak, it also mentions that Codex plans to build a small electronic pet and place it on the Codex desktop to indicate the status of the current session.

This electronic pet has 8 preset avatar appearances, and users can also create and use their own virtual avatars.

Source:

Another leak says that OpenAI is developing an intelligent agent for ChatGPT (code name Hermes). It includes an agent builder, templates, scheduling, options for using agents in Slack, adding applications, skills, files, memory, instructions, and more.

Source:

Right now, Codex is an actively developed product, so OpenAI definitely won’t give the local agent product market to Claude.

Not to mention, a few days ago, even Gemini—another company led by the AI industry big brother OpenAI—quietly released a desktop app, but it was widely criticized by a bunch of netizens as “a complete disaster.”

All we can do is encourage OpenAI and Gemini to quickly end Claude’s lead in local agent assistants and coding.

The whole world has long suffered from Company A.