Codex’s latest replace frequency has been insane.
Over the previous two months, OpenAI has added one thing new to Codex virtually each few days.
First got here plugins, built-in browsers, pc operations, PR critiques, distant SSH, and cellular entry… Then on Might 21, Codex joined “Loopy Thursday,” rolling out a number of main updates without delay: one-click display screen sharing with Codex, enabling Codex to give attention to a purpose for prolonged intervals, continued distant utilization after display screen lock, and help for team-shared plugins and utilization knowledge monitoring.
Beforehand, there was a extensively circulated meme: Get up, and see one other Claude replace. Now, Codex is not any completely different.
Nevertheless, Claude has been up to date extra regularly and with finer enhancements, whereas Codex has launched extra main options.
Notably, they’re each up to date in the identical path—enterprise portals and real-world workflows.
Claude Code has already demonstrated the worth of this path, and Anthropic has even begun convincing the market that cutting-edge mannequin corporations don’t should burn cash indefinitely—additionally they have the potential to attain profitability.
Codex is doing the identical factor, backed by OpenAI, which is making ready for its IPO at the moment.
ChatGPT has demonstrated that OpenAI has customers, however customers don’t equal enterprise, and recognition doesn’t essentially translate to revenue. Particularly for a cutting-edge mannequin firm, the prices of compute energy, coaching funding, and inference bills are substantial. OpenAI must show to the market that it’s not simply able to creating viral chatbots, however may combine AI into manufacturing workflows that enterprises are genuinely prepared to pay for.
Codex’s frequent updates are filling on this hole.
It isn’t only a improvement instrument—it is OpenAI’s clearest present path to demonstrating industrial worth.
01
What has Codex been doing over the previous two months?
We created a picture utilizing ChatGPT Photographs 2.0 to point out what updates Codex has remodeled the previous two months.
On March 24, search and settings are synchronized.
The Codex App now consists of historic thread search and fast navigation to latest threads, whereas synchronizing key settings between the Codex App and the VS Code extension. This can be a foundational expertise enchancment designed to assist customers shortly resume earlier duties and guarantee a extra constant expertise throughout the desktop and editor.
On March 25, the plugin system was launched.
Codex now helps plugins. Plugins can package deal expertise, software integrations, and MCP server configurations to reuse workflows, and are supported by Codex App, CLI, and IDE extensions.
On April 9, the code evaluation workflow was enhanced.
The Codex App has added collapsible inline evaluation feedback, completely different evaluation modes, Git summaries, and supply blocks. Codex is now diving deeper into code evaluation and PR collaboration.
On April 12, file and terminal context enhancements have been applied.
Codex has added file search to the command menu, helps previewing photos, PDFs, and Markdown within the sidebar, introduces a terminal tab for every thread, and permits customers to instantly ask Codex questions after deciding on textual content.
April 16: Codex for nearly the whole lot.
That is the primary main milestone up to now two months: OpenAI is remodeling Codex right into a extra complete AI workstation. This replace consists of an built-in browser, pc operations, thread automation, a activity sidebar, PR workflows, consequence previews, SSH distant connections, multi-terminal and multi-window help, Intel Mac compatibility, and a collection of recent plugins.
April 23, automated approval evaluation.
Codex can first route eligible approval requests to an automatic evaluation agent to evaluate threat, then show the evaluation standing and threat stage, permitting the consumer to determine whether or not to approve.
On Might 5, the Codex entry token was launched.
Workspace house owners and directors for ChatGPT Enterprise can permit members to create Codex entry tokens to be used in trusted, non-interactive native workflows akin to scripts, schedulers, and personal CI runners. Codex is more and more integrating with CI, automation, and enterprise engineering methods.
On Might 7, Codex entered Chrome.
Codex has launched a Chrome extension that allows parallel work inside browser tabs with out instantly taking on the consumer’s browser. Customers may management which web sites are permitted to make use of Codex. As browsers function the entry level for a lot of backend methods, inside instruments, and net debugging eventualities, this step brings Codex nearer to real-world workplace environments.
On Might 14, Codex added cellular management help.
OpenAI permits customers to make use of Codex by the ChatGPT cellular app by connecting to a Mac operating the Codex app; customers may view activity progress, approve actions, examine code diffs, and see take a look at outcomes on their cellular units. This replace additionally consists of the overall availability of Hooks, entry tokens, and enterprise admin setup guides. Codex is starting to operate as a distant workflow agent.
Might 21: Appshots, Goal Mode, Lock Display screen Distant Entry, and Plugin Sharing.
That is the second main function. Appshots can ship screenshots of the present Mac window and out there textual content on to Codex; the Goal Mode has formally launched, permitting customers to provide Codex a purpose and have it work constantly on that purpose for hours and even days; and Lock Display screen Distant Utilization permits Codex to proceed working desktop purposes even after the Mac is locked, eliminating the necessity to “maintain a line open.”
On the similar time, ChatGPT Enterprise now helps team-shared plugins; the built-in browser’s annotation capabilities have been additional enhanced to permit direct changes to font, coloration, spacing, and different kinds.
The options themselves are definitely essential, however the total replace tendencies are equally value noting. Whether or not it’s Appshots, goal modes, Chrome extensions, entry tokens, or plugin sharing, all of those are addressing the basic necessities for getting into real-world workflows: seeing the state of affairs in actual time, driving duties ahead, and managing dangers successfully.
To see the complete image, what must be accomplished is contextual understanding.
Actual improvement duties not often happen solely inside a code editor. File search, file preview, terminal tabs, built-in browsers, browser annotations, Chrome extensions, and Appshots all essentially cut back the associated fee for customers to explain context to AI.
Beforehand, you needed to inform AI the place the issue was by describing it or utilizing Ctrl+C/V; now, OpenAI needs Codex to see this stuff instantly.
To maneuver duties ahead, long-running duties and distant execution capabilities are important.
The goal mode addresses the query of “whether or not it may be sustained.” Distant entry on cellular units and distant utilization whereas locked permit duties to proceed progressing even when customers are away from their computer systems. Entry tokens and hooks additional combine Codex into enterprise engineering methods akin to scripts, schedulers, and CI runners.
Managing threat is as much as the corporate and its crew.
For particular person builders, instruments are judged totally on usability, however enterprise instruments contain much more complicated points: find out how to handle permissions, find out how to distribute plugins, who’s utilizing them and the way a lot, find out how to conduct threat critiques, whether or not they can combine with CI, and whether or not they are often centrally managed by the crew.
Codex has additionally accomplished intensive work on this space. The plugin system permits workflows to be packaged and reused; plugin sharing permits groups to uniformly distribute instruments; automated approval critiques mitigate dangers related to agent execution; and entry tokens and enterprise admin settings combine Codex with present engineering and governance processes inside the enterprise.
02
The hope of all the village
The replace to Codex has introduced it very spectacular consumer progress.
In early March, Codex had round 1.6 million weekly lively customers. By Might 14, OpenAI formally talked about that over 4 million individuals have been utilizing Codex every week. Which means that inside roughly two months, Codex’s weekly lively customers greater than doubled.
This progress trajectory can’t be separated from the underlying mannequin’s capabilities; customers are prepared to entrust Codex with actual duties extra regularly provided that it might probably really get the job accomplished. Particularly after GPT-5.5, Codex has gained a stronger basis in coding, instrument utilization, long-context dealing with, and multi-step activity execution.
However having a mannequin alone is not sufficient—the market will not pay simply because a mannequin’s benchmark improves; it cares extra about whether or not these capabilities can generate income.
That is additionally one thing OpenAI should make clear earlier than going public.
OpenAI holds many playing cards, however every card comes with its personal uncertainty.
ChatGPT is the biggest consumer gateway, demonstrating OpenAI’s skill to draw world customers and consumer-grade subscriptions. Nevertheless, the bigger the consumer base, the higher the inference prices; whether or not consumer-grade subscriptions can maintain long-term funding by a cutting-edge mannequin firm nonetheless wants additional validation.
API is a basic income supply, enabling the sale of mannequin capabilities to builders and enterprises. Nevertheless, the API market is extremely prone to cost competitors, and enterprise prospects will not be locked right into a single mannequin supplier. The extra general-purpose the mannequin functionality, the extra doubtless prospects are to make use of a number of fashions together.
ChatGPT Enterprise, Brokers, and {industry} options signify OpenAI’s direct entry into the enterprise market. Nevertheless, for these merchandise to really combine into enterprise workflows, they require time, gross sales efforts, integration, and industry-specific deployment.
Additional out, OpenAI additionally has {hardware}, knowledge facilities, multi-cloud partnerships, and computing infrastructure. These tales are extremely imaginative but additionally heavier, farther eliminated, and extra capital-intensive. They help the long-term imaginative and prescient however are tough to instantly tie to short-term enterprise returns.
The industrial worth of Codex is less complicated to clarify, as its audience is clearly outlined: builders and engineering groups.
These are people who find themselves already prepared to pay for providers. Engineers’ time is pricey, software program initiatives take a very long time, and code upkeep carries excessive prices. Each step—bug fixes, testing, code critiques—might be quantified by way of price.
Software program improvement can be one of the vital important manufacturing processes for any enterprise. Monetary corporations depend on threat administration and buying and selling methods, retail corporations on provide chain and membership methods, healthcare corporations on knowledge and compliance methods, and media corporations on content material administration and distribution platforms. Even non-tech corporations require intensive inside instruments, knowledge pipelines, automation scripts, and enterprise methods to function… In the present day, just about each firm will depend on software program methods.
In different phrases, Codex integrates into the areas the place companies spend cash and devour human assets day by day.
In a way, it’s OpenAI’s hope for crafting a compelling IPO narrative. This has turn out to be particularly essential as OpenAI prepares to enter the capital markets.
Within the itemizing narrative, OpenAI is not going through questions like “Does AI have a future?” The true problem is that this: Can a cutting-edge mannequin firm discover a clear, secure, and sufficiently worthwhile enterprise path past huge computational investments?
Extra troubling is that Anthropic has already taken a step additional on this challenge.
03
Anthropic has taken the lead.
Codex have to be delivered to the forefront for one more important purpose: Anthropic, one among OpenAI’s greatest rivals, has already cast a path within the enterprise area.
Though OpenAI nonetheless leads by way of income, in keeping with The Info, OpenAI’s income within the first quarter of 2026 was roughly $5.7 billion, increased than Anthropic’s $4.8 billion throughout the identical interval. However the challenge now’s not simply how a lot income they generate—the actual stress on frontier mannequin corporations is whether or not their income progress can outpace their price progress.
OpenAI’s first-quarter income was excessive, however its adjusted working margin was roughly -122%. On this foundation, for each greenback of income, adjusted working prices could quantity to about $2.22, leading to a lack of $1.22.
Over the previous few years, outsiders have questioned whether or not giant mannequin corporations are burning by an excessive amount of money—coaching, inference, GPUs, and expertise prices all signify bottomless pits. The extra customers there are, the extra frequent the calls, and the heavier the prices turn out to be.
Anthropic’s latest alerts have expanded the probabilities surrounding this matter.
In line with The Wall Road Journal, Anthropic expects income to exceed $10.9 billion within the second quarter of 2026 and to strategy its first quarterly working revenue, with an estimated working revenue of roughly $559 million.
Whereas this doesn’t imply Anthropic has utterly solved its cash-burning drawback, it sends an important sign to the market: frontier mannequin corporations don’t should rely indefinitely on funding—as soon as their fashions are highly effective sufficient and their merchandise carefully align with high-value enterprise use instances, income progress can outpace prices.
Anthropic doesn’t have a mass-market entry level like ChatGPT, nor does it pursue as many concurrent narratives. Its path is narrower and extra centered: instantly concentrating on high-value use instances the place enterprises are prepared to pay—significantly in areas akin to builders, finance, regulation, analysis, knowledge evaluation, and inside information work.
Claude Code is among the most consultant examples. It initially emerged as vital instrument for builders, centered on programming eventualities, and steadily expanded to incorporate duties, plugins, permissions, crew administration, and enterprise governance, turning into a key entry level for Anthropic into enterprise workflows. Builders undertake it first, groups observe swimsuit, and finally it turns into a part of company procurement and budgeting.
In April 2026, Anthropic’s adoption fee amongst Ramp’s pattern enterprises rose to 34.4%, whereas OpenAI’s dropped to 32.3%. Though this knowledge is predicated solely on enterprise spending patterns on the Ramp platform and doesn’t signify a full market survey, it at the least means that Anthropic is gaining momentum in enterprise paid use instances.
The stress on Codex lies right here.
OpenAI’s income scale nonetheless leads the {industry}, but when it goals to enter the capital market, it can’t rely solely on consumer base or mannequin capabilities. It wants a product nearer to enterprise manufacturing environments to display that it might probably flip AI into secure company income.
If Claude Code proves that developer workflows can function Anthropic’s enterprise entry level, then Codex should show that OpenAI may succeed on this path.
Codex CEO Tibo Sottiaux lately half-jokingly summarized the corporate’s “grasp plan”: launch higher, extra environment friendly fashions, launch improved merchandise weekly, purchase extra compute energy, and spend extra time scrolling on X.
A greater mannequin determines whether or not Codex can actually get work accomplished; extra frequent product updates decide whether or not Codex can combine into actual workflows; extra computing energy determines whether or not all of this may help more and more giant utilization.
These are all essential for itemizing.
In different phrases, Codex’s latest flurry of updates isn’t nearly catching up on options—it’s additionally about following the enterprise path that Anthropic has already cast.
ChatGPT has demonstrated that OpenAI has customers.
And Codex goals to show that OpenAI is a worthwhile enterprise.
This text is from the WeChat public account “Letter AI,” authored by Yuan Xinyue.