By Marcus, Sales, HUMAN HIVE
Every AI model your company builds on has a shutdown date, and several big ones land between now and April 2027. The tech fix is a migration; the bigger risk is human: nobody owns the list, and the habits people built around one model quietly stop working.
The 60-second version
OpenAI, Google, Microsoft, Amazon and Anthropic all publish shutdown dates for the AI models behind business tools, and several land between Nov 19, 2026 and Apr 14, 2027.
The same model can stop on different days depending on where you bought it: o3 ends Nov 19 on Microsoft's cloud and Dec 11 on OpenAI's.
Name one owner, list every AI tool your team uses and the model under it, and ask each vendor for their date this week.
What happened
On October 14, 2026, OpenAI told developers that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini leave its API on January 14, 2027. MIXED reported that this leaves 92 days between notice and switch-off, while OpenAI's own page promises "at least 6 months" for generally available models.
That is one date in a crowded calendar. Here is what the makers' own pages say, read on November 4:
OpenAI: Agent Builder, the Evals platform and reusable prompts shut down Nov 30, 2026. GPT-5, GPT-5 mini, GPT-5 nano, GPT-5 pro, o3 and o3-pro shut down Dec 11, 2026.
Anthropic: Claude Sonnet 4.5 retires on the Claude API Nov 30, 2026, with Claude Sonnet 5.5 as the recommended replacement.
Google's Gemini API: from Dec 1, 2026, requests to gemini-2.5-pro and gemini-2.5-flash automatically route to newer models, billed at the new models' rates.
Microsoft (Azure): o3 retires Nov 19, 2026, three weeks before OpenAI's own date. Two of the three GPT-4o versions, GPT-4o mini and GPT-4.1 mini run to Apr 14, 2027, three months after OpenAI's. Microsoft's policy page is blunt: "Retirement dates aren't extendable."
Amazon Bedrock: 26 third-party models, including Llama, DeepSeek, Gemma, Mistral and Qwen models, reach end of life Mar 30, 2027 in the listed regions, and Amazon says migration "will not happen automatically."
One honest limit: these pages cover developer platforms, the models built into company systems and vendor products. They do not announce changes to the ChatGPT, Gemini or Claude apps people open in a browser.
Why it matters: the human side
A shutdown date is a technical event. The damage is human.
Nobody owns the list. Model names hide inside vendor products, scripts and workflows somebody set up last spring. If no single person knows which models your tools run on, nobody sees the date coming.
The switch can be silent. Google's Dec 1 change routes requests to a different model rather than failing. Microsoft's standard deployments auto-upgrade, while its provisioned deployments do not. Either way, the answer quality, tone and cost your people got used to can change without anyone deciding it should.
Training goes stale on a date nobody chose. Teams learn a tool's habits: how to phrase a request, what to double-check, where it tends to drift. When the model underneath changes, some of that know-how resets. People blame themselves, or the tool, and quietly stop using it.
What Mitch has taught on this
Mitch has been saying for a while that chasing the newest model misses the point. In a video on which AI model to pick, he put it this way: if every company stopped building new models today, "you couldn't master it all in five, six, seven years." His advice was simple: "Learn how to use it effectively. Learn how to use it efficiently."
And for anyone who feels behind every time a new name appears: "You're not late. You're not behind. You're right on time."
That is the real lesson of shutdown season. Models come and go on someone else's calendar. The skill of working well with them, asking clearly, checking the output, knowing when it is wrong, is what carries over. That part belongs to your people.
Best practices: five moves for this week
Name one owner. The COO or head of operations assigns one person (not a committee) to own the AI tool list. Their job is the calendar, not the code.
Build the list in 30 minutes. Columns: tool, team that uses it, vendor, model underneath, where it was bought (direct, Azure, Amazon, Google), shutdown date. Blank cells are your to-do list.
Ask every vendor one question. "Which model runs this product, where is it hosted, and what is your plan for its retirement date?" Put the answer in writing.
Re-test what people rely on. Before each switch, have the people who use the tool every day run their five most common requests on the new model and compare. They will spot drift faster than any dashboard.
Refresh training, not just settings. When a model changes, L&D sends a one-page "what's different" note and holds a 15-minute huddle so people adjust together instead of struggling alone.
What's driving people crazy this week
Mixed signals. In a public Google developer forum thread from July, developers reported that gemini-2.5-pro returned a "no longer available to new users" error while the lifecycle page still listed a later date. The reply said access was limited to earlier users. The human fix: never plan on a date you read once. Your owner rechecks the maker's page monthly and writes down what it says that day.
What this means for the next 1 to 3 years
Model churn is not slowing down. We expect three shifts. First, "AI tool owner" becomes a real job line, the way someone already owns software licenses. Second, smart buyers start writing model-change notice into vendor contracts, so a silent swap becomes a breach. Third, the best L&D teams stop teaching a tool and start teaching a method: how to brief any model, test it and catch it when it drifts. Tools expire. Methods travel.
How we evolve
The human skill to build here is adaptability: staying calm and curious when the ground moves. It matters more as AI gets better, because better models arrive faster, and every arrival is a small change your people have to absorb. Mitch makes the same point in his ADAPT! keynote with the BlackBerry story: the market leader lost because the world moved faster than it did. One daily practice: at the end of each day, write one line about something that changed at work and how you responded. In a month, you will see your own pattern, and you can choose a better one.
Try this today (under 30 minutes)
Open a shared sheet with the six columns above and fill in the first three tools you can name.
Ask at your next team meeting: "Which AI tool would hurt most if it changed tomorrow, and who would notice first?"
Send one vendor the one-question email from step 3 of the best practices.
Free tool: the vendor question, ready to copy.
"Hi [name], quick planning question. Which AI model runs [product] for us, where is it hosted, and what is your plan for that model's retirement date? We are building our 2027 calendar and want to stay ahead of any changes. Thank you."
FAQ
Will ChatGPT or Gemini stop working for my team?
These dates cover the developer platforms behind business tools, not the consumer apps. The apps change on their own schedules, so check each vendor directly.
Why does the same model have different shutdown dates?
Each cloud sets its own schedule. o3 ends Nov 19, 2026 on Microsoft Azure and Dec 11, 2026 on OpenAI's API. Claude Sonnet 4.5 retires on Anthropic's API Nov 30, 2026 but runs on Amazon Bedrock to Apr 8, 2027.
What is the first thing a leader should do?
Name one owner for your AI tool list this week. Everything else follows from someone knowing what you use. For a quick read on where your people stand, try our free readiness assessment.
Tools change. Adoption is a behavior problem. Find your team's friction points in one short conversation.