The pace of iteration in the AI field is refreshing outside perceptions. OpenAI officially announced on October 14 that GPT-5.5 has been retired from ChatGPT, ChatGPT Work, and the account-login Codex, although the related API interfaces are currently unaffected. At the same time, specified versions of multiple legacy models, including GPT-4o, have entered a countdown to removal and will be officially decommissioned on January 14, 2027, leaving developers and enterprise users only about three months to migrate.
With the full launch of new-generation models such as GPT-6.1, significant improvements have been achieved in both response speed and factual accuracy, while invocation prices are also more market-competitive. Currently, ChatGPT's weekly active user base has surpassed the 1.2 billion mark, and how to balance compute investment between serving such a massive user base's inference needs and intensive model training has become a practical challenge facing OpenAI.
Amid fierce market competition, OpenAI and Anthropic are racing comprehensively on model capability, response speed, and invocation cost. The two sides' recent密集 iterations—although bringing users stronger performance—have also repeatedly exposed service anomalies, once again highlighting the critical importance of underlying infrastructure construction and operational stability.
Regarding the retirement of these legacy models, the official side has not only provided practical alternative model options but also specifically reminded users to promptly check and update the model specification settings of related old tasks. Developers and users are advised to flexibly choose the most suitable model and reasoning effort based on specific business scenarios and actual task requirements, thereby achieving the most efficient resource utilization in real-world applications.