Your AI Model Is Being Retired: Why It Keeps Happening and What to Do (2026)
OpenAI has replaced GPT-5.5 with the GPT-5.6 family, Anthropic has retired the original Claude 4 models, and Google keeps sunsetting older Gemini versions. Here's why AI companies retire models so often, what it means for you, and the simple steps to take.
TL;DR. AI companies retire old models regularly. In mid-2026 alone, OpenAI replaced GPT-5.5 with the GPT-5.6 family (and retired GPT-4.5 earlier in the year), Anthropic retired the original Claude 4 models, and Google keeps sunsetting older Gemini versions. For most chat users this is an automatic, usually-better upgrade — nothing to do. If you use the API or built model-specific workflows, switch before the cutoff or things break.
If you've ever opened your AI app to find "your" model gone, you've hit one of the quieter realities of this era: models get retired on a schedule. Here's what's happening now, why, and the few steps that matter.
What's happening right now
- OpenAI replaced GPT-5.5 with the GPT-5.6 family (Sol, Terra, Luna) on July 9, 2026. Earlier in the year it retired GPT-4.5 from ChatGPT (June 27, moving users to GPT-5), and o3 is scheduled to leave ChatGPT on August 26, 2026. When a model string is retired, API calls to it return an error, so developers need to switch first.
- Anthropic retired its older Claude models. The original Claude 4 models (Sonnet 4, Opus 4) and Haiku 3.5 have been retired on most platforms (they linger on Amazon Bedrock and Google Cloud), Opus 4.1 is deprecated, and Opus 4.7 fast mode is being removed on July 24, 2026. The current lineup is Sonnet 5 (default), Opus 4.8, Haiku 4.5, and Fable 5.
- Google is deprecating older Gemini models on a rolling basis — older 2.0 and preview models in mid-2026, with Gemini 2.5 versions extended to no earlier than October 16, 2026. Google keeps a public deprecations page with dates and recommended replacements.
This isn't a glitch or a downgrade — it's routine lifecycle management, and it will keep happening.
Why models get retired
Three reasons, all mundane:
- Newer models are better and cheaper. When a new generation beats the old one on quality, speed, and cost, keeping the old one around mostly adds confusion and expense.
- Maintaining many versions is costly. Every live model is infrastructure to run, secure, and support. Companies prune.
- It pushes everyone onto the upgrade. For the typical user, being moved off a retired model is effectively a free step up.
The trade-off is churn: when you're moved to a new model, its behavior can shift slightly — tone, formatting, how it handles an edge case.
What it means for you
If you use the chat app (ChatGPT, Gemini, Claude): usually nothing to do. You'll be moved to a newer default automatically, and it's typically an improvement. Your chats, Projects, and files aren't deleted — only the model behind them changes.
If you use the API or built workflows: this is the group that gets bitten. Calls to a retired model string will fail. Update to a supported model before the cutoff date.
Either way, re-check anything you tuned to a specific model. If you have a saved prompt, a custom Project, or an automation that depended on one model's exact quirks, give it a quick test after the switch — a new model may respond a little differently.
What to do (quick checklist)
- Note the dates for tools you rely on (e.g., o3 → August 26, Opus 4.7 fast mode → July 24; check Google's deprecations page for Gemini).
- Check your default model in each app's settings so you know what you're being moved to.
- API/dev users: swap retired model strings for supported ones before the cutoff.
- Re-test tuned prompts and workflows on the new model.
- Don't over-attach to a version. Build prompts and processes that work on "a good current model," not on one specific release.
The bigger pattern
Model retirements are now a permanent feature of using AI, not a one-off. The healthiest posture is to treat the specific model as a detail and your workflow as the thing that lasts — so the next retirement is a non-event. If you're choosing among current options, see which Claude model you should use, and for the broader landscape start at the AI Basics hub.
Sources
- OpenAI: Retiring GPT-4o and older models
- OpenAI Help Center: ChatGPT release notes
- Google: Gemini API deprecations
- TechRadar: OpenAI retires the last of the GPT-4 models
See Claude set up for your job
Skip the theory — pick your profession and get the real workflows, ready-to-use prompts, and exact setup for your work.
See Claude set up for your job
Real workflows and ready-to-use prompts, profession by profession.
Frequently asked questions
Which AI models are being retired in 2026?+
It's a moving target. OpenAI retired GPT-4.5 from ChatGPT on June 27, 2026 (users were moved to GPT-5), and on July 9, 2026 it replaced GPT-5.5 with the GPT-5.6 family; o3 is scheduled to leave ChatGPT on August 26, 2026. Anthropic retired the original Claude 4 models (Sonnet 4, Opus 4) and Haiku 3.5 on most platforms, deprecated Opus 4.1, and is removing Opus 4.7 fast mode on July 24, 2026. Google continues to deprecate older Gemini versions on a rolling basis. When a model string is retired, API calls to it start returning an error, so developers need to switch first.
Are Gemini models being retired too?+
Yes. Google regularly deprecates older Gemini versions — for example, older 2.0 and preview image models were scheduled to shut down in mid-2026, and Gemini 2.5 versions were extended to no earlier than October 2026. Google publishes a deprecations page so you can see dates and recommended replacements.
What happens to me when my AI model is retired?+
If you use the consumer chat app, usually nothing you need to do — you're automatically moved to a newer, typically better model. If you use the API or built workflows that depend on a specific model, you must switch to a supported model before the cutoff date, or your calls will fail.
Why do AI companies retire models so often?+
Newer models supersede older ones on quality, speed, and cost, and maintaining many versions is expensive. Retiring old models pushes everyone onto better, cheaper ones — so for most users a retirement is effectively a free upgrade. The trade-off is churn: behavior can shift slightly when you're moved to a new model.
Should I be worried about losing my work?+
Your chats, Projects, and files aren't deleted when a model is retired — only the underlying model changes. The main thing to re-check is any prompt or workflow you tuned to a specific model's quirks, since a new model may respond a little differently.
Related Guides
NotebookLM Is Now Gemini Notebook: What Changed, What's New, and What to Do
Google renamed NotebookLM to Gemini Notebook on July 16, 2026. Your notebooks are safe — nothing moved or broke. But there are real new features for Pro and Ultra users, and Search integration is coming. Here's what actually changed.
AI Skills Are Now Required in 3 Out of 4 Tech Job Postings: What Employers Are Actually Asking For
Dice analyzed 7 million U.S. tech job postings: AI skill requirements jumped from 15% in January 2024 to 75% in June 2026. Here's exactly which skills employers are listing — and what non-technical professionals should do about it.
Google Started Training AI on Your Search Photos and Voice Recordings — Here's the Opt-Out
In June 2026, Google quietly expanded what it trains AI on — now including images from Google Lens, voice recordings from Search and Translate, and uploaded files. Here's what changed, which services are affected, and the two-step opt-out that most guides miss.