OpenAI plans to phase out GPT-5, GPT-4o and other older models: Know more

OpenAI said it will remove access to GPT-5, GPT-4o and several older models from ChatGPT on February 13, citing low usage and a shift toward newer versions

OpenAI has announced that it will retire several older AI models from ChatGPT, including GPT-5 (Instant and Thinking), GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini. According to the company, these models will stop being available in ChatGPT from February 13, 2026. OpenAI said the decision was driven by low usage, as most users have already moved to newer models. GPT-4o, which had briefly returned after being retired during the GPT-5 rollout, will now be phased out again.

OpenAI retires older GPT models

According to OpenAI, the changes currently apply only to ChatGPT and will not affect the API at this stage. The company said most users are now using GPT-5.2, while only around 0.1 per cent of daily users still rely on GPT-4o. Retiring older models allows OpenAI to focus its resources on improving the versions that see the most real-world use. 

OpenAI added that user feedback has played a key role in shaping its newer models. Based on how people use ChatGPT, the company said it has improved conversational quality, creative support, and customisation options such as tone and style controls. It has also worked on reducing responses that feel overly cautious or preachy. In addition, OpenAI said it is developing a version of ChatGPT intended specifically for adult users, while using age-prediction tools in most markets to better manage how younger users interact with the service. 

GPT-4o: Background

Among the models being retired, GPT-4o has had a more complex history. After being discontinued during the initial GPT-5 rollout, the model was temporarily restored following strong user feedback. Some Plus and Pro users said they relied on GPT-4o for creative ideation and preferred its conversational style, which led OpenAI to bring it back while working on improvements. 

GPT-4o has also faced criticism over what OpenAI described as “sycophantic” behaviour. In this context, sycophancy refers to the model’s tendency to agree with users even when they are incorrect, rather than prioritising factual accuracy. OpenAI rolled back an update to GPT-4o in April 2025 after users flagged this issue, reverting it to an earlier version with more balanced responses. 

  OpenAI CEO Sam Altman has previously acknowledged underestimating how emotionally attached some users were to GPT-4o, particularly during the transition to GPT-5. Despite this, OpenAI said the decision to retire the model was not taken lightly and reflects a broader shift toward newer models that better align with how most users now interact with ChatGPT.