Skip to content

GuidesAI at work

Custom GPTs in your company: what to do before 11 December 2026

Custom GPTs in your company: retired on 11 December 2026, migration to a plugin, ChatGPT Business workspace agents, prices and limits.

9 min read

A company no longer really creates a custom GPT in November 2026: OpenAI retires GPTs on 11 December 2026, on 11 February 2027 for customers granted an extension, and Free, Go, Plus and Pro accounts can already no longer create them. The useful question has turned around, since what matters now is what becomes of the GPT the team has used for a year, and where to build the next one: in a plugin for individual use, or in a ChatGPT Business workspace agent for an assistant the whole team shares.

This guide follows OpenAI’s help centre as it reads on 2 November 2026. It describes how a GPT migrates to a plugin, what gets lost on the way, how to build a workspace agent, what the Business plan costs and what none of these pieces settles on your behalf. The other ways to create an agent, in n8n, in Claude or by adopting an AI employee, are covered in our guide to creating an AI agent.

Can you still create a custom GPT in 2026?

On a personal account, no: since the retirement was announced, Free, Go, Plus and Pro accounts can no longer create or publish a new GPT, so subscribing to Plus does not reopen that door. In company workspaces, OpenAI sets the end of creation through a notice shown in the workspace, and it ended on 26 October 2026 in the Enterprise workspaces concerned. As of 2 November 2026, ChatGPT’s French pricing page still lists custom GPTs in the Plus plan, which the help centre contradicts.

The retirement applies to every GPT, the ones a team built as well as the ones it uses from the GPT Store. A GPT was never an agent in the strict sense: you put it together in conversation or in a configuration tab, with instructions, at most twenty knowledge files of 512 MB each, capabilities such as web search or the code interpreter, and either connected apps or custom actions, never both. It answered when someone wrote to it, and that is precisely what its successors go beyond.

One point deserves checking before anything else: which account were your GPTs created on? A GPT built by an employee on a personal Plus subscription belongs to that account, not to the company, and it will leave with the person before it even leaves on 11 December. This is the most common form of AI used without a tool the company provided, and the inventory of GPTs is often its first written trace.

What happens to an existing GPT on 11 December 2026?

OpenAI migrates GPTs to plugins, available on every plan, which bundle skills, reference files and connected apps. The migration starts from the My GPTs page with the Migrate to plugin button: the GPT’s instructions become a skill, its knowledge files become reference files, and its connected apps carry over.

Two things do not make the trip. Custom actions, those calls to an API described in OpenAPI that let a GPT write to a CRM or to Notion, for example, have to be rebuilt, possibly as an MCP server, and the model chosen for the GPT is not kept. Once the migration is done, the original GPT becomes read-only and the plugin starts out private, which means sharing it again with the colleagues who used it.

Migrating a team GPT, step by step

Take a ten-person communications agency that built a “Client brief” GPT last year: it turns meeting notes into a brief following the agency’s method, draws on three reference documents and drops the result into Notion through a custom action. The example is built for illustration, but it brings together the three cases a migration meets: what carries over as it is, what carries over transformed and what does not carry over at all.

Start with the inventory: the list of GPTs people actually use, their owner, the account they live on and the actions they call. Next, copy each GPT’s instructions into a company document before launching anything, because they are often the only written record of the method they describe. Then run the migration from My GPTs, and check that the skill really carries the instructions and that the reference files are the current versions.

Finally, rebuild the action to Notion, either through the Notion app if it is offered or through an MCP server, and run the plugin on about twenty briefs that were already written and whose right result you know, three times each. A plugin that produces three different briefs from the same notes is not ready, however good each one is, and that is the method set out in our protocol for testing an agent. Only share it after that.

How do you create a workspace agent in ChatGPT Business?

In ChatGPT Business, Enterprise and Edu, the successor to the team GPT is called a workspace agent: a shared agent that runs in the cloud, has a memory, can be scheduled or triggered through an API, and answers in ChatGPT as well as in Slack. You create one from the Agents tab by describing the work to be done, then add apps such as Google Drive, Slack or SharePoint, custom MCP servers, skills and files.

For each app, the creator chooses between two ways of connecting: with the account of the person using the agent, or with an account that belongs to the agent. The first choice means everyone sees through the agent only what they can already see, while the second gives the agent fixed permissions, whoever is talking to it. In the second case OpenAI recommends a service account limited to what is strictly needed, and it is the same question that an agent’s identity raises: should two people with different permissions get the same answer?

Write actions are set to “always ask” by default, and action constraints let you, for example, allow emails to be sent only to a given domain. The agent can be shared privately, by link or in the organization’s directory. Since 29 September 2026, the Business plan has added team tasks, which carry neither each person’s memory nor their personal instructions, and the ability to call ChatGPT in Slack and Microsoft Teams.

How much does ChatGPT Business cost, and what do you pay on top?

As of 2 November 2026, the ChatGPT Business Standard seat costs $20 per person per month on annual billing and $25 monthly, the Premium seat $100 and $125, with two seats minimum. OpenAI shows no official price in euros on its page, and the estimates resellers publish vary too much to be repeated here.

What you pay for How
Standard seat $20 a month on annual billing, $25 monthly, two seats minimum
Premium seat $100 a month on annual billing, $125 monthly
Workspace agents in credits, beyond what each seat includes
Plugins included, on every plan

For a ten-person agency, the Standard subscription comes to $200 a month on annual billing before the agents’ credits, which depend on how much work they are given. Data in a Business workspace is not used to train models by default, but data residency in Europe is open only to new Enterprise and Edu customers, and not to Business. A company whose clients require hosting in Europe therefore has to look at the Enterprise plan, or at another tool.

What the workspace agent leaves unsettled

A workspace agent fixes much of what the GPT lacked, memory, scheduled runs and a presence in Slack, and leaves three questions untouched that the company has to settle itself. None of them comes down to the model, and all of them come up by the second month.

The first is approval. “always ask” is a default setting, and the agent’s creator can switch some write actions to “never ask”, often on the day the confirmation starts slowing them down. The rule we apply is not a setting: what leaves the company goes through a person, what stays inside does not need to, and the rule is only worth something if it is built into the tool rather than written into the instructions, as our article on what an agent should not be able to do explains.

The second is memory. Each agent keeps its own, so the quotes agent, the briefs agent and the support agent each know a version of the same client, and those versions drift apart without ever saying so. Two agents live side by side without harm when their domains do not overlap, which is rarely the case in a small business, where the same clients run through every subject.

The third is the timetable. The retirement of GPTs, announced a few months before the date, shows that an agent configured with a vendor lives at the pace of the vendor’s decisions, and what was built for one model or format often breaks with the next. Keep your instructions and your reference files in a company document, not only in the tool, so that the next migration starts with a copy rather than a reconstruction.

What Balt changes for a team that had GPTs

Balt is an AI employee of the company rather than an assistant per person: you write to him in Teams, Slack, WhatsApp, on the phone or in his web app, and what he learns with one person serves the whole team. A GPT’s instructions become, with him, a skill that he writes himself as he is corrected and that can be read back in the console, once for the whole company, and the work that comes back becomes a routine that runs at a set time or on an event.

Nothing he prepares for a client or a supplier goes out without your approval, and that is not a setting anyone can loosen. He works in more than 3,200 connected tools, costs from €99 a month for 40,000 credits shared by the whole team, with no limit on users and 20,000 credits free at sign-up, and his database is hosted in France and the European Union on every plan. For open-ended writing, deep research or code, ChatGPT stays broader than Balt, and the full comparison is on our ChatGPT Business or Balt page.

Where do you start before 11 December?

With the inventory, this week: which GPTs are really used, which account they live on, who owns them and which actions they call. Then copy every set of instructions into a company document, migrate what serves a single person to a plugin, and keep the rebuild as a shared agent for what the whole team uses, deciding before you start who approves what goes out. What comes next is chosen with the order of delegation that works, starting with the task whose result you can check in ten seconds.

Frequently asked questions

  • Can you still create a custom GPT?

    Not on a personal account: as of 2 November 2026, Free, Go, Plus and Pro accounts can no longer create GPTs, and creation ended on 26 October 2026 in the Enterprise workspaces concerned. In other company workspaces, the notice shown in the workspace sets the date. Every GPT is retired on 11 December 2026, or on 11 February 2027 for customers granted an extension.

  • What replaces the custom GPT?

    Two things, depending on the use. The plugin, available on every plan, bundles skills, reference files and apps, and it is what existing GPTs migrate to. The workspace agent, reserved for ChatGPT Business, Enterprise and Edu, is an agent the team shares, with a memory, a schedule and the ability to answer in Slack.

  • What do you lose when you migrate a GPT to a plugin?

    Custom actions, meaning calls to an API described in OpenAPI, and the choice of model do not carry over: they have to be rebuilt, for example with an MCP server. The original GPT becomes read-only once migrated, and the plugin starts out private, so it has to be shared again with the colleagues who used it.

  • How much does ChatGPT Business cost?

    As of 2 November 2026, the Standard seat costs $20 per person per month on annual billing and $25 monthly, the Premium seat $100 and $125, with two seats minimum. OpenAI publishes no official price in euros on its page, and workspace agents consume credits beyond what each seat includes.

Sources

  1. OpenAI, Custom GPT retirement and migration FAQ (accessed 2 November 2026)help.openai.com
  2. OpenAI, Creating and editing GPTs (accessed 2 November 2026)help.openai.com
  3. OpenAI, Plugins in ChatGPT (accessed 2 November 2026)help.openai.com
  4. OpenAI, ChatGPT Workspace Agents for Enterprise and Businesshelp.openai.com
  5. OpenAI, Introducing workspace agents in ChatGPT (22 April 2026)openai.com
  6. OpenAI, ChatGPT Business release notes (29 September 2026)help.openai.com
  7. OpenAI, ChatGPT Business pricing, French page (accessed 2 November 2026)openai.com
  8. OpenAI, What is ChatGPT Business?help.openai.com
  9. OpenAI, Data residency for ChatGPThelp.openai.com
  10. OpenAI, ChatGPT pricing for individuals, French page (accessed 2 November 2026)chatgpt.com

Balt does this work

  • SkillsWhat Balt learns about the way you work.
  • RoutinesThe work that comes back, handled every week.

Read next

Give Balt its first job.

Thirty seconds to add it to Teams and get to work.

Try for free