OpenAI Sets Retirement Dates for o3 and GPT-4.5 in ChatGPT

OpenAI has confirmed sunset dates for two of the more notable models in its ChatGPT lineup. GPT-4.5 will retire from ChatGPT on June 27, 2026, after a 30-day sunset period. The reasoning model o3 follows on August 26, 2026, after a 90-day sunset window. Both are currently visible only to paid ChatGPT users through the model picker.

The move is part of OpenAI's ongoing push to consolidate its consumer-facing model lineup around GPT-5.5 Instant — the new default — plus a tighter set of specialized reasoning models. API customers are not affected by these specific retirements, but the writing is on the wall for both models across all OpenAI surfaces.

Context: Why OpenAI Keeps Pruning Its Lineup

OpenAI has been aggressive about retiring older models throughout 2026. GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini were all queued for retirement earlier this year, leaving a much simpler menu of choices for paid users.

The strategic reason is straightforward: serving multiple legacy models at scale costs real money, fragments the user experience, and slows down focus on the newer flagship architectures. With GPT-5.5 Instant now powering most ChatGPT traffic as the default, keeping older models around mostly serves a small slice of power users who developed workflows around specific model behaviors.

There's also a quality argument. GPT-4.5 was OpenAI's attempt at a maximalist non-reasoning model — large, fluent, expensive to run, but not as strong as newer mixture-of-experts and reasoning-equipped successors. o3 was a milestone reasoning model when it launched, but the o-series has since advanced considerably and newer reasoning models cost less to serve.

Details: What's Actually Changing

GPT-4.5 retirement on June 27. After the 30-day sunset window ends, GPT-4.5 will no longer appear in the ChatGPT model picker for Plus, Pro, Team, or Enterprise users. The model picker will continue to offer GPT-5.5 Instant as the default plus specialized reasoning options.

o3 retirement on August 26. The longer 90-day sunset reflects o3's heavier use among power users who built reasoning-heavy workflows. After that date, o3 disappears from ChatGPT. OpenAI is directing users to the newer o-series reasoning models, which it says outperform o3 on most benchmarks at a lower cost.

API access. Critically, these changes apply only to ChatGPT. API customers can still access both models programmatically, though OpenAI has separately published deprecation timelines for older API models that affect developers. Companies running production workflows on o3 or GPT-4.5 via the API have more time to plan migrations, but should still expect API sunsets within 6–12 months.

Industry Impact: Lineup Hygiene as Strategy

The ChatGPT model picker has become an interesting product-strategy lens. OpenAI's bet is that fewer, better-marketed models drive more usage than a long menu of variants. Anthropic has made a similar bet with Claude Opus, Sonnet, and Haiku families. Google has consolidated around Gemini 3.5 Flash and Pro tiers.

The trade-off is that retiring popular models inevitably frustrates the loyal users who depend on them. Every retirement cycle brings complaints from developers, content creators, and prompt engineers who had tuned specific workflows to specific model behaviors. OpenAI's typical playbook is to ship a new flagship that compensates with broad capability gains, then retire older models a few months later.

For enterprises, the lesson is to avoid hard-coding any model name into long-running workflows. Use abstraction layers — a single internal model alias that maps to the current best model — so retirements become a routing change rather than an emergency rebuild.

Expert Perspectives

Developers in the OpenAI community have largely accepted that frequent retirements are now part of the platform. Power users who relied on o3 for specific reasoning tasks have noted that newer o-series models behave differently — sometimes better, occasionally worse on niche workloads — which means the 90-day sunset is essential testing time, not a buffer to ignore.

Industry analysts have pointed out that OpenAI's retirement pace is faster than enterprise customers prefer. Some compliance-heavy industries need model stability for validation work. Anthropic's longer model support windows and its "managed model" enterprise tiers are now being marketed as a differentiator.

What's Next

If you're a ChatGPT user, three things to do. First, audit any saved Custom GPTs or workflows pinned to GPT-4.5 or o3 and re-test them against GPT-5.5 Instant or the newer o-series options. Second, if you have a prompt library you've optimized for those models, plan to refactor before the relevant sunset date. Third, watch for OpenAI's typical practice of announcing replacement models that "match or exceed" the retiring model on benchmarks — those are usually the right migration targets.

For API customers, OpenAI's official deprecations page is the canonical source. Don't rely on ChatGPT release notes for API timelines.

Bottom Line

OpenAI is continuing to thin out its model lineup, with GPT-4.5 gone in less than a month and o3 retired by the end of August. The retirements clear shelf space for newer, cheaper-to-serve models and tighten the user experience around GPT-5.5 Instant. For most users, the migration is straightforward. For the power users and developers who built workflows around the retiring models, the next 30 to 90 days are the migration window — and waiting until the sunset deadline is the path to broken workflows.