ChatGPT-4o
❗This article is a stub. You can help by expanding it.
#appeals channel in either Zulip or Discord to request removal.An article may be flagged as a stub when it is missing major elements needed to make it useful to a reader. You can help by adding missing sections, verifiable sources, relevant company policies and communications, etc. to make the article more complete.
ChatGPT-4o was a language model that belonged to the ChatGPT family, that became controversial due to safety problems, AI psychosis and attempts to remove it from OpenAI's platform.
Background
[edit | edit source]The model was announced and released on May 13, 2024, as a flagship model capable of "reasoning through audio, images and text in real time". [1]
Sycophancy
[edit | edit source]ChatGPT-4o was highlighted for its sycophantic responses, that made the chatbot to reply in an acommodating way, using a lot of compliments and agreeing with the user, even if the prompt contains info that may harm the user or someone else. [citation needed]. This brought consequences as dependency, with several users claiming they have formed a "romantic relationship" with the AI, chatbot psychosis, and in the most severe cases with users committing suicide. [2] An update launched in 2025 worsened this behavior in the model. When users noticed this, OpenAI temporarily removed the model to review the issue. [3]
Attempt to remove the model
[edit | edit source]On 07 August 2025, GPT-5 rolled out as the newest model available for use. Along with this, legacy models were removed, including GPT-4o. The removal of 4o caused an outrage amongst the users of this model. This, combined with the complaints of GPT-5 performance,[4] made OpenAI to bring back legacy models, but behind a paywall, so only Plus users were able to access to it. [5]
Definitive removal
[edit | edit source]On 29 January 2026, OpenAI announced they will be removing definitely for non-API users some legacy models, including GPT-4o. The model was officially removed on 13 February 2026. ChatGPT-4o will be removed for API usage on April 2026.[citation needed]
OpenAI's response
[edit | edit source]OpenAI claimed "only a 0.1 percent of users were using ChatGPT 4o".
Lawsuit
[edit | edit source]
Consumer response
[edit | edit source]- A lot of users complained in social media about the removal of the model. Some users that used to have a strong dependency of the model were severely psychologically affected by OpenAI's decision. [citation needed]
- A Change.org petition was created to bring back the model. The petition has surpassed 21,000 signs as on February 16 2026.
- Some users started developing their own AI models with outputs of ChatGPT-4o or clones that behave similarly. [citation needed]
- Several users who analysed and researched concluded OpenAI's claim of 0,1% was false and it is estimated a 5-35% of Plus users were actual consumers of GPT-4o.[citation needed]
References
[edit | edit source]- ↑ OpenAI (13 May 2024). "Hello, GPT-4o". OpenAI. Archived from the original on 16 Dec 2025. Retrieved 15 Feb 2026.
- ↑ Hill, Kashmir (2025-06-13). "They Asked an A.I. Chatbot Questions. The Answers Sent Them Spiraling". The New York Times. Archived from the original on 2025-10-13. Retrieved 2025-02-15.
- ↑ Franzen, Carl (30 Apr 2025). "OpenAI rolls back ChatGPT's sycophancy and explains what went wrong". VentureBeat. Archived from the original on 15 Aug 2025. Retrieved 15 Feb 2026.
- ↑ Edwards, Benj (11 Aug 2025). "The GPT-5 rollout has been a big mess". Ars Technica. Archived from the original on 9 Jan 2026. Retrieved 5 Feb 2026.
- ↑ Franzen, Carl (8 Aug 2025). "OpenAI returns old models to ChatGPT as Sam Altman admits 'bumpy' GPT-5 rollout". VentureBeat. Archived from the original on 2026-02-05. Retrieved 5 Feb 2026.