South Florida · Mexico City The OGs of digital marketing · now fluent in AI ● AI systems in production
Journal · AI Marketing

HighLevel Is Retiring GPT-4.1 in Conversation AI: How to Switch Your Agents Safely

HighLevel now recommends GPT-6 Sol for Conversation AI agents and says it will retire GPT-4.1, so any agent still on GPT-4.1 should be moved and tested before that happens. HighLevel’s changelog gives no retirement date; OpenAI’s API shutdown for GPT-4.1 is 14 January 2027.

This is what the changelog says, what it leaves out, and a short way to switch an agent without surprising a customer.

What did HighLevel actually announce?

HighLevel’s changelog has an entry titled “GPT-6 Sol Is Now the Recommended Model in Conversation AI.” It makes four changes for prompt-based Conversation AI agents. GPT-6 Sol is listed first in the Model menu with a Recommended tag. New prompt-based agents now start on it by default. Existing agents keep the model you selected. And GPT-4.1 is on its way out: the entry says OpenAI is deprecating it and that HighLevel “will retire it in Conversation AI.”

The entry also says agents that used GPT-5 or GPT-5 Mini already run on GPT-6 Sol, with reply generation up to 4.5× faster, and that those agents need no action. The only group that has to do something is the one still on GPT-4.1.

How much weight should the 98.4% figure carry?

HighLevel reports a 98.4% pass rate for GPT-6 Sol on its own Conversation AI evaluation, against 96.3% for GPT-4.1, at about the same cost per reply. It also gives a typical generation time of 1.68 seconds and 2.19 seconds for the slowest 5% of replies.

Read that as a vendor’s test of its own product. The changelog does not say what the evaluation contains, how many conversations it covers or what counts as a pass, so you cannot tell whether it resembles your agent’s job. A booking agent for a dental clinic and a lead-qualifying agent for a yacht broker fail in different ways. The 2.1-point gap is a reason to try the new model, not proof that your agent will improve. HighLevel says the same thing in its own notes: each model writes in its own way, so test before you save.

When does GPT-4.1 actually go away?

Here is the catch. HighLevel’s page, as we opened it on 19 November 2026, shows no retirement date and no publish date. We are not going to supply one. The only dated source for the underlying model is OpenAI’s deprecations page, and it covers OpenAI’s API, not HighLevel’s product:

ModelOpenAI noticeOpenAI API shutdownHighLevel says
GPT-4.114 Oct 202614 Jan 2027Will be retired in Conversation AI; no date given
GPT-511 Jun 202611 Dec 2026Deprecated; agents already run on GPT-6 Sol
GPT-5 Mini11 Jun 202611 Dec 2026Deprecated; agents already run on GPT-6 Sol

HighLevel can retire a model in its own product on a different day from OpenAI’s API shutdown, so the January date is a ceiling to plan around, not HighLevel’s deadline. The safe reading is simple: if an agent is still on GPT-4.1, move it on your schedule now rather than on someone else’s.

How do you switch an agent without breaking it?

HighLevel’s own steps are short. Add a test pass before you save, because that is where problems show up:

  1. Open AI Agents → Conversation AI and open the agent.
  2. On the Build tab, open the Model menu and choose GPT 6 Sol.
  3. Run your list of real customer messages through Test your agent before saving (see the next section for what belongs on it).
  4. Click Save, then watch the first live conversations in the agent’s logs.
  5. Keep a note of the previous setting, so you can see what changed if a reply goes wrong.

Do one agent at a time, starting with the one that talks to the fewest customers.

What belongs on your test list?

Because each model writes differently, a switch can change tone, length and how an agent handles an edge case, even when the prompt is untouched. Build the list from conversations your agent has already handled:

  • The five most common questions, with the answers your business actually wants given.
  • A booking request, end to end, including a time that is not available.
  • A question the agent should not answer (price promises, legal or medical advice, anything outside its brief), to confirm it still hands off.
  • A message in Spanish and one in English, if you serve both, to check the language stays right.
  • An angry or confused customer, to check it hands the conversation to a person.

Save the replies side by side with the old ones. The aim is not a perfect score; it is to see differences before a customer does.

What should an agency with many accounts do?

The risk here is the quiet one. Because existing agents keep the model they were set to, nothing breaks until the retirement happens, and then a client’s agent may be moved without anyone having read its replies. So take an inventory now: list every sub-account, every Conversation AI agent in it and the model each one uses. Mark the ones on GPT-4.1, rank them by how many customer conversations they handle, and book the switches in that order.

Also check the scope. HighLevel says its recommendation applies to prompt-based Conversation AI agents, so do not assume the same note covers other agent types. The changelog’s 18 November entry for a new Managed Agents API, which lets you create, update and publish Agent Studio agents programmatically, is a separate product and does not change the Conversation AI note. If you want help setting up the agents themselves, see our GoHighLevel systems and AI agentic marketing pages, and for the broader picture of how we apply AI to a business, AI marketing.

Related reading: if your agents answer customers on WhatsApp, the WhatsApp messages API deadline is another change to put on the same calendar, and can AI agents use your website covers the other direction, AI agents acting on your site.

Want your business found on Google and in ChatGPT? Ciao Bella engagements start at $5,000 a month, no setup fee. See how we work.

Verified 19 November 2026 against HighLevel’s changelog entry and OpenAI’s API deprecations page, both opened that day

Building on the web since 1999 · South Florida and Mexico City

Frequently asked questions

Do I have to change my Conversation AI agents right now?

Only if an agent is on GPT-4.1. HighLevel says existing agents keep the model you selected, and that agents that used GPT-5 or GPT-5 Mini already run on GPT-6 Sol. HighLevel gives no retirement date for GPT-4.1, so switch on your own schedule and test first.

Will GPT-6 Sol cost more per reply than GPT-4.1?

HighLevel says it costs about the same per reply. It gives no exact pricing in the changelog entry, so check your own usage after the switch.

What is the retirement date for GPT-4.1 in HighLevel?

HighLevel’s changelog entry does not give one. OpenAI’s deprecations page lists 14 January 2027 as the API shutdown date for GPT-4.1, but HighLevel may retire it in its own product on a different day.

Does this apply to every type of HighLevel AI agent?

The changelog says the recommendation applies to prompt-based Conversation AI agents. It does not say the same applies to other agent types, so check each one rather than assuming.

How do I switch an agent to GPT-6 Sol?

Open AI Agents, then Conversation AI, then the agent. On the Build tab choose GPT 6 Sol from the Model menu, test it in Test your agent, and click Save.

Sources

  1. HighLevel Changelog — GPT-6 Sol Is Now the Recommended Model in Conversation AI Opened 19 November 2026. Source for the recommendation, the 98.4% and 96.3% pass rates, reply times, the GPT-5 note, the GPT-4.1 retirement and the switching steps. The page shows no publish date.
  2. HighLevel Changelog — recent entries Opened 19 November 2026. Used to check that no dated retirement notice for GPT-4.1 appeared among the latest entries.
  3. OpenAI — Deprecations Opened 19 November 2026. GPT-4.1: notice 14 October 2026, API shutdown 14 January 2027. GPT-5 and GPT-5 Mini: notice 11 June 2026, shutdown 11 December 2026.
  4. HighLevel Changelog — Introducing Managed Agents API Opened 19 November 2026, listed under 18 November. Shows that the Managed Agents API (Agent Studio) is a separate release from the Conversation AI model note.

Running AI agents on HighLevel?

Tell us what you have set up — we’ll tell you what to check before the old model goes.

Want to see what it costs first? Engagements start at $5,000 a month. See how we work.

We use your details only to reply and talk about working together. We never sell them. Privacy policy.

Free newsletter

Get our AI-search notes by email.

One short take a day: what changed in Google and ChatGPT, and what to do about it. Unsubscribe anytime.

No spam. We've been online since 1999. We know what spam looks like.

Hey, want to chat?
Scroll to Top