OpenAI has set a date for the retirement of GPT-5.5 in consumer products: on October 14, 2026, the model will disappear from the selector in ChatGPT, ChatGPT Work, and Codex across all plans — consumer, Business, Enterprise, and Edu. Access to GPT-5.5 via the OpenAI API remains, and product users are offered to switch in advance to its successors — GPT-5.6 Sol or GPT-6 Astra.

image

What happened

The announcement took place on September 15, 2026: the official ChatGPT account on X published a farewell post, "Thanks for everything, 5.5," which received 2.2 million views and 30 thousand likes, and a note about the model's removal from products appeared on OpenAI's release notes page. The retirement itself is scheduled for October 14, 2026: from this date, GPT-5.5 will disappear from the model selector in ChatGPT, ChatGPT Work, and Codex across all plans, including consumer, Business, Enterprise, and Edu. Codex users who log in via a ChatGPT subscription are officially recommended by OpenAI to replace GPT-5.5 with GPT-5.6 Sol or GPT-6 Astra. This concerns only the model choice in product interfaces, not the model's availability for developers.

Context

The rotation reveals two different lifecycle policies that OpenAI is pursuing in parallel: in subscription products, model versions are replaced quickly and on schedule, while via the OpenAI API, previous versions continue to live as a long tail for developers. Notably, the successors were released before the retirement date: GPT-5.6 Sol and GPT-6 Astra are already available, so until October 14, all three models exist in the products simultaneously — a rare window for controlled comparison on your own tasks, limited in time. Less than a month remains from the September 15 announcement to the retirement. The materials themselves contain no benchmarks, quality comparisons, or explanation of the reasons: the decision is framed as a planned removal of the version from the consumer selector, not as a rejection of the model as such.

Why this matters for the industry

For the industry, the main signal is not the fact of the retirement itself, but the splitting of access channels: the model disappears from subscription interfaces but remains in the OpenAI API, meaning it behaves as a swappable, configurable dependency. Operational work falls to teams that have integrated GPT-5.5 into Codex and ChatGPT Work workflows via subscription: by October 14, they need to find all scenarios using this model, choose between GPT-5.6 Sol and GPT-6 Astra, and run regression tests on their own cases. Code API integrations do not need to be changed — it is enough to ensure that they do not rely on subscription quotas, while API costs are billed separately from the subscription. A less obvious consequence: tutorials, prompt engineering, and evaluations calibrated for GPT-5.5 through product interfaces lose validity after the rotation, so a reproducible access point to a specific version remains fixing the model via the API.

Why this matters for users

An ordinary user does not need to do anything before October 14: in ChatGPT, ChatGPT Work, and Codex, everything will continue to work as before. After this date, GPT-5.5 will disappear from the model selector, so it is better to switch to GPT-5.6 Sol or GPT-6 Astra in advance and manage to compare both successors on your own tasks before the rotation takes place. If you specifically need GPT-5.5 for a long time, access via an API key will remain, but it is paid for separately from the ChatGPT subscription.

What is still unknown / limitations

The available materials contain no data on the quality of GPT-5.5 and its successors: no benchmarks, no comparisons, no explanation of the reasons for the retirement, so the claim that the model is inferior to its successors is not supported by anything. The API guarantee is also limited: sources confirm the preservation of access at the current moment, but the lifespan of GPT-5.5 in the API and the order of future retirements on the API side are not disclosed, so the idea of the API as a more stable contract than a subscription is plausible but remains an interpretation. The expectation that GPT-5.5 will remain a legacy API model for a long time is a forecast, not a confirmed fact.

Sources

Author

Look at AI, editorial team