AI News

ChatGPT and Codex: What We Know

Direct Answer

ChatGPT and Codex are clearly converging, but OpenAI has not announced one fully merged app. Corey McClain's video documents a redesigned ChatGPT web sidebar, updated project organization, Codex-like themes and profile controls, and third-party reports of a "web merge." Those observations fit OpenAI's broader direction: the company already describes Codex as available inside the ChatGPT desktop app, and several agentic products share one usage pool.

That does not confirm the most dramatic claims. OpenAI has not said that Chat, Work, and Codex will become one automatic agent, that regular Chat will lose its separate usage treatment, or that a $500-$600 "Pro Max" plan powered by Cerebras will launch at DevDay.

The useful conclusion

Expect fewer boundaries between OpenAI's interfaces and more shared infrastructure behind agentic work. Do not make subscription, migration, or budget decisions from strings found in client code. The confirmed action is to audit and migrate any custom GPTs you need before their scheduled retirement.

Watch the Analysis

Credit and evidence: the interface walkthrough and speculation come from Corey McClain, published on 25 September 2026. DevDay timing, Codex access, usage pools, credits, pricing, and custom-GPT migration were checked against current official OpenAI pages on 26 September 2026.

An Evidence Ladder for the Claims

ClaimStatusWhat supports it
DevDay is on September 29ConfirmedOpenAI's official DevDay 2026 page
Custom GPTs are retiringConfirmedOpenAI's migration FAQ sets December 11, 2026 as the standard retirement date
Codex and Work can share usageConfirmedOpenAI documents a shared agentic allowance for Codex, Work, Word, Excel, and PowerPoint on eligible plans
Regular Chat remains outside that poolConfirmed for the current systemOpenAI explicitly says regular Chat usage is not included
ChatGPT web is receiving a redesigned interfaceCreator-observed rolloutMcClain's account shows the new sidebar, project views, settings, and themes
ChatGPT and Codex will become one automatic agentUnconfirmedInference from interface convergence and third-party reports
A Pro Max tier will cost $500 or $600UnconfirmedReported strings and screenshots, not an official product or pricing announcement
Pro Max will use Cerebras and deliver extreme speedUnconfirmedThird-party speculation; no official OpenAI product page confirms it

This distinction matters because interface experiments can ship gradually, change names, remain account-specific, or never become a public product. Client code may contain tests, placeholders, regional prices, annual equivalents, abandoned options, or internal labels.

The Interface Is Converging

McClain shows a ChatGPT web experience that looks more like the desktop app. Scheduled tasks receive a dedicated view instead of appearing only like conversations. The overflow menu groups Projects, Health, Finances, Sites, Maps, and GPTs. Settings expose ChatGPT and Codex visual themes, while profile controls resemble the desktop experience.

The strongest official corroboration is structural rather than visual. OpenAI's current help center repeatedly refers to using Codex inside the ChatGPT desktop app. The company also supports Codex across the app, web, IDE extension, CLI, and cloud with the same ChatGPT login. That is already a unified account and multi-surface product family, even if Chat and Codex retain distinct modes.

Interface merge is not capability merge.

A common navigation system, theme, profile, or billing page does not prove that conversational chat, knowledge work, and code execution will use identical models, permissions, sandboxes, or limits.

Projects Are Becoming the Organizing Layer

The most immediately useful change in the video is mundane and overdue: projects and conversations show recency, projects can be pinned, and individual conversations can be pinned within a project. That makes ongoing work easier to resume without relying on the global conversation list.

A good project still needs more than a folder name. Keep a short operating record:

  • Goal: the outcome and the user or customer it serves.
  • Source of truth: approved files, repositories, data systems, and owners.
  • Permissions: what the agent may read, edit, publish, buy, or message.
  • Decision log: accepted choices, rejected approaches, and open questions.
  • Completion evidence: tests, screenshots, reports, links, and human approval.

Better navigation helps you find the work. It does not replace project context or review controls.

Custom GPTs Are Retiring: Migrate Carefully

The transcript briefly says "skills are going away," but the official announcement says the opposite transition: custom GPTs are retiring and creators are being guided toward plugins. Under the migration workflow, a GPT's instructions become a skill inside the new plugin.

OpenAI schedules standard retirement for 11 December 2026. A limited Enterprise deferral can extend qualifying workspaces to 11 February 2027. Migration availability and notices may arrive at different times, so follow the account or workspace where the GPT was created.

Custom GPT elementMigration behaviorRequired review
Published instructionsBecome a plugin skillCheck trigger description, ordering, exceptions, and output format
Knowledge filesBecome plugin reference filesCheck freshness, permissions, sensitive data, and retrieval quality
Connected appsAdded as plugin apps where supportedReauthorize accounts and verify action approvals
Custom actionsDo not transfer automaticallyReplace with a supported app or reviewed custom MCP server
Selected modelDoes not carry overRetest quality, cost, latency, and tool behavior
Existing conversationsDo not migrateExport any decisions or examples you still need
Sharing and public accessDo not carry overTest privately, then use the separate plugin-sharing or submission process
  1. Inventory the GPTs you created and the public GPTs your workflows depend on.
  2. Publish essential drafts before migration; the migrator uses the latest published version.
  3. Save five familiar test prompts, one difficult case, and the expected result for each GPT.
  4. Document every custom action and the external account or secret it uses.
  5. Migrate one low-risk GPT first and compare behavior side by side.
  6. Ask an intended user to install and test the private replacement before wider sharing.

What Usage Actually Shares Today

The video's most important question is whether Chat and Codex will draw from one meter. OpenAI's current documentation gives a more precise baseline:

  • On eligible Plus and Pro plans, Codex, ChatGPT Work, Word, Excel, and PowerPoint can share the same agentic allowance.
  • Purchased credits can fund eligible activity after included usage is exhausted.
  • Regular Chat usage is not included in the Work and Codex agentic pool.
  • Images, voice, file uploads, and other surfaces can have separate limits and reset periods.
  • Actual consumption varies by model, task complexity, context, tools, execution location, and speed.

So one of McClain's possible futures already resembles the current system: the products can share a front door and an agentic pool while ordinary Chat remains distinct. DevDay could change presentation or entitlements, but there is no official basis yet for assuming regular chats will consume the same credits as long-running Codex work.

Budget agentic work by completed task.

Track the model, time, credits, retries, tool calls, output quality, and human review required. A faster response is valuable only if it produces a usable result with fewer correction loops.

The Pro Max Story Is Still Speculation

McClain reviews third-party posts that claim client code contains "Pro Max" labels and $500 or $600 price points. He also discusses a possible Cerebras-backed speed tier. Those clues are interesting, but they cannot establish a launch, final price, included usage, region, tax treatment, infrastructure partner, or performance.

As of 26 September, OpenAI's public ChatGPT pricing page lists Free, Go, Plus, Pro, Business, and Enterprise. It does not list Pro Max. Its official credit guidance says purchase amounts, prices, payment methods, spending limits, and availability can vary by account, region, and plan.

If a premium speed tier appears, evaluate it with the work you already run:

  1. Choose ten repeatable tasks with known acceptance criteria.
  2. Run the same models, permissions, and source material on the current and new tier.
  3. Measure wall-clock time, agentic credits, failures, retries, and reviewer minutes.
  4. Calculate value from time saved on accepted outputs, not advertised tokens per second.
  5. Keep the lower tier unless the monthly benefit consistently exceeds the price difference.

What to Watch at DevDay

OpenAI's official DevDay page confirms only the date and location: 29 September in San Francisco. It calls the event the company's biggest of the year but does not promise the launches discussed in the video.

QuestionEvidence that would answer it
Is the product actually unified?An official product name, navigation model, availability list, and migration path
Will Chat and agentic work share usage?A plan table that explicitly includes regular Chat in the agentic pool
Is Pro Max real?An OpenAI pricing or help page with price, region, entitlements, and limits
Is faster inference a paid tier?Named models, latency terms, quality expectations, and availability
How do plugins replace GPTs?Migration dates, permission model, public distribution, analytics, and action support
Will the new web interface reach everyone?A release note with plans, platforms, regions, and rollout timing

What to Do Before the Announcements

  1. Do not buy around a rumor. Wait for the official plan table and your own account's upgrade screen.
  2. Audit custom GPTs now. Identify owners, instructions, files, actions, users, and sensitive connections.
  3. Capture current limits. Save the usage page and reset windows so you can compare any post-DevDay change.
  4. Record a baseline. Time five real Codex or Work tasks and note credits, retries, and quality.
  5. Separate navigation from governance. A prettier shared interface should not broaden permissions or remove approvals.
  6. Read the release notes after the keynote. Product pages and help articles define what you can actually use.

Video Chapters

TimeTopic
00:00New interface features
01:39Project organization and pinning
02:24Custom GPT migration
04:16Platform and interface convergence
07:26Usage-limit scenarios
09:51Pro Max tier speculation
14:31Conclusion and DevDay watch

Sources

This article records the evidence available on 26 September 2026, before DevDay. Interface rollouts, migration dates, plan entitlements, and pricing can change; use the official pages and your account notice for decisions.

Common questions

Has OpenAI announced one combined ChatGPT and Codex app?
No. OpenAI already delivers Codex through the ChatGPT desktop app and several surfaces, and the interfaces are converging, but it has not officially announced that ChatGPT and Codex are becoming one product or that their names and modes will disappear.
Are ChatGPT and Codex usage limits combined?
OpenAI currently says Codex, ChatGPT Work, Word, Excel, and PowerPoint can share an agentic allowance or credit pool on eligible plans. Regular Chat usage is documented separately and is not included in that agentic pool.
Are custom GPTs going away?
Yes. OpenAI says custom GPTs are scheduled to retire on December 11, 2026, with a limited deferral for qualifying Enterprise workspaces. Migration access and plan-specific timing can vary, so follow the notice in the account where the GPT was created.
Does migrating a custom GPT preserve everything?
No. Instructions become a skill, knowledge files become plugin reference files, and connected apps become plugin apps. Existing conversations, selected models, sharing settings, and custom actions do not transfer automatically.
Has OpenAI confirmed a ChatGPT Pro Max plan?
No. The video discusses third-party reports of plan names and prices found in client code, but OpenAI had not listed or announced a Pro Max plan on its official pricing pages as of September 26, 2026.
When is OpenAI DevDay 2026?
OpenAI officially scheduled DevDay 2026 for September 29 in San Francisco. The event page does not promise a unified ChatGPT and Codex app, a Pro Max subscription, or any specific launch covered by the rumors.
Share
X LinkedIn Reddit
Build Yours

Want a system
like this one?

Book a free 30-minute call. We map your situation, identify the highest-impact automation, and figure out if we are a fit.

Book Free 30-min Call