AI This Week: Custom GPTs Get an Expiry Date (Sept 29, 2026)

Published September 29, 2026 • By Rujal Tuladhar

If your team built a custom GPT last year to answer customer questions, write quotes or onboard new staff, it now has an expiry date. That is the change from the past two weeks most likely to cost a GTA business real time if nobody notices it, and it arrived quietly in a help-centre article.

It was not the only one. AI model prices fell again, Amazon let sellers run their store from a chat window, Google started testing an assistant that phones local businesses on behalf of customers, and Microsoft changed the default billing on new Copilot Business licences. Below is what changed between September 11 and 29, what it costs, and who should actually do something about it. Every fact comes from a page we opened this week; the opinions are ours.

Custom GPTs stop running Dec 11

They move to plugins; actions and model choice do not carry over

GPT-6 Sol: $2 / $10 per million tokens

Half the old price; Luna is $0.10 / $0.50

Claude Opus 5.5 costs 40% less to run

Anthropic's claim for typical workloads vs Opus 5

Amazon Seller Central inside Claude

US beta only; Canadian stores not supported yet

Gemini now phones businesses

Testing with US Pixel 11 owners

Copilot Business billing default changes Nov 2

New CSP licences get usage billing, $10/user default cap

AI This Week: Custom GPTs Get an Expiry Date (Sept 29, 2026) at a glance — the short version.

Custom GPTs are being retired: check yours before December 11

OpenAI plans to retire custom GPTs across ChatGPT plans, with December 11, 2026 set as the retirement date, although Virtualization Review notes that OpenAI says timing can vary by account and workspace. The replacement is plugins. A built-in migration turns a GPT's instructions into a skill inside a new plugin, copies its knowledge files across as reference files, and can bring connected apps along.

The catches are what matter. According to the same report, existing GPT conversations do not migrate, the model you picked does not carry over, and custom actions may have to be rebuilt with an available app or a custom MCP server. Sharing resets too: migrated plugins start private, and people who used your GPT do not automatically get access to the replacement. Mixed News reports that personal accounts (Free, Go, Plus and Pro) already cannot create or publish new GPTs, while existing ones keep working for now.

For most small businesses the GPT itself is simple to move. The risky ones are those wired to something else: a GPT that looks up stock in a spreadsheet, books into a calendar, or posts into a CRM through a custom action. Those are the ones that will quietly stop working if nobody rebuilds them.

What it means for you: List every custom GPT your team relies on this week, flag any that use custom actions, and migrate and retest them in October, not the week of December 11.

Running AI in your own tools just got cheaper, again

On September 22 OpenAI released GPT-6 Sol and GPT-6 Luna to developers. Artificial Analysis lists Sol at $2 per million input tokens and $10 per million output tokens, down from $4 and $20 for GPT-5.6 Sol, and Luna at $0.10 and $0.50, down from $0.20 and $1.20. Cached input, which is the text an automation re-sends every time it runs, like your instructions and price list, now gets a 90% discount on reads. The Cryptonomist notes that OpenAI confirmed to VentureBeat these are permanent prices, not a promotion.

The same day Anthropic released Claude Opus 5.5 at $4 per million input tokens and $20 per million output, 20% below Opus 5, with cache reads cut to $0.20. Anthropic says it will cost 40% less than Opus 5 on typical workloads at default settings, and it is included in the Pro, Max, Team and Enterprise plans.

What this means in practice: if you pay a developer or an automation tool per use, the model part of that bill should be falling. A job that reads every incoming email and drafts a reply was already cheap; on Luna-class pricing, the model cost for a few thousand emails a month is small enough that the real costs are setup and someone checking the output.

What it means for you: If a vendor charges you per AI task, ask whether they have moved to the new models and whether the price cut reaches you.

Amazon puts Seller Central inside Claude, but not for Canada yet

At its Accelerate conference on September 23, Amazon launched a Selling Partner plugin that brings Seller Central into Claude and into Amazon's own Quick assistant. Epinium's breakdown lists what it can touch: listing prices, titles and images, 7-, 30- and 60-day inventory forecasts, FBA shipments, and sales and listing-health analytics. Changes need the seller's approval, with an extra confirmation for permanent or costly actions. Enterprise DNA adds that sellers can set workflows either to surface recommendations or to take approved actions automatically.

Now the limits. Per Epinium it is a US-only beta for sellers on the Professional plan, it works with one account and one marketplace at a time, it leaves out advertising and margin advice, and Canada and Europe are not supported, with international expansion planned but not dated. Both sources report that primary account holders can get 12 months of Amazon Quick Plus free, with the offer running to December 31, 2026.

What it means for you: Amazon.ca-only sellers should wait; if you also sell on Amazon.com, join the beta and start with inventory forecasts before you let it change prices.

Your phone may soon ring with an AI on the line

On September 24 TechCrunch reported that Google is testing a feature that lets Gemini place phone calls for its users: asking a store whether something is in stock, booking a restaurant, moving an appointment, or putting items on hold. The customer follows a live transcript and can take over at any time. It starts with US Pixel 11 owners who pay for a Gemini subscription.

It is US-only today, but it tells you where customer calls are heading. The report does not say how Gemini identifies itself to the business it calls. For a Toronto salon, clinic or hardware store, the practical point is that some future callers will be software reading from your website and Google Business Profile, and it will act on whatever those say.

What it means for you: Check that your hours, holiday closures and booking rules on Google Business Profile match reality, and tell front-desk staff to treat AI callers like any other customer.

Housekeeping: Gemini apps, Copilot billing and DevDay

Gemini connects to Squarespace, Zoho and more. On September 23 Google began rolling out 14 new connected apps for the Gemini app, including Airtable, monday.com, PandaDoc, Zoho, Adobe, Squarespace and Webflow. You connect them in Gemini settings or by typing @ in a chat. Neither Google's post nor Android Authority's write-up lists which countries get each one, so check your own settings before planning around it.
Copilot Business billing changes November 2. From that date, new Microsoft 365 Copilot Business licences bought through a CSP partner come with usage-based billing switched on. Stoneridge Software reports a default cap of $10 per user per month for eligible AI services, and that existing licences are not affected. Topedia notes that new pay-as-you-go agents and services are enabled by default. Microsoft's cost-management documentation shows where admins set spending limits and turn off the auto-apply setting.
OpenAI DevDay is today. The keynote is at 1 p.m. Eastern, livestreamed free. Superpower Daily notes the public schedule names no products; rumours of a $500 plan are unconfirmed. We will cover anything that matters for small businesses next week.

What it means for you: If you buy Copilot through a reseller, ask them before November 2 what spending cap your new licences will carry and who gets the alerts.

The bottom line

Two weeks ago the story was AI assistants reaching into your accounting and CRM. This fortnight it is about upkeep: a tool you built last year gets a retirement date, the prices underneath your automations drop, and your billing defaults change unless someone looks. None of it is hard, but all of it is easy to miss when you are running a business. That is the part we handle at Nova Toronto: AI automation that we build, watch and move when the platforms shift, including custom GPT migrations before the December deadline. If you want a second opinion on what you have running, the first consultation is free.

Want this working in your business?

Book a free 30-minute call and we'll map out which of these would actually save you time and win you customers.

Book a Free Consultation

Sources: Virtualization Review: OpenAI to Retire Custom GPTs, Replace Them with Plugins (Sept 28, 2026) • Mixed News: OpenAI is retiring custom GPTs, here's what carries over to plugins (Sept 18, 2026) • Artificial Analysis: GPT-6 Sol and Luna push the cost efficiency frontier (Sept 22, 2026) • The Cryptonomist: OpenAI GPT-6 pricing slashes costs with new Sol and Luna models (Sept 23, 2026) • Anthropic: Claude Opus 5.5 (Sept 22, 2026) • Epinium: Amazon Accelerate 2026 - Seller Central in Claude (Sept 26, 2026) • Enterprise DNA: Amazon Lets Sellers Run Their Business Through Claude (Sept 2026) • TechCrunch: Google tests letting Gemini make phone calls, initially for US Pixel owners (Sept 24, 2026) • Google: A new wave of Connected Apps is rolling out to Gemini (Sept 23, 2026) • Android Authority: Gemini just got new connected apps for work, creativity, and life (Sept 23, 2026) • Stoneridge Software: Automatic Usage-Based Billing Is Coming to Microsoft 365 Copilot Business (Sept 21, 2026) • Topedia: New Microsoft 365 Copilot Business subscriptions from CSP partners will come with PAYG preconfigured (Sept 20, 2026) • Microsoft Learn: Usage-based billing and cost management for Copilot Credits (updated Sept 25, 2026) • Superpower Daily: OpenAI DevDay is tomorrow - when and how to watch (Sept 28, 2026)