OpenAI removed GPT-4o from ChatGPT today, ending access to one of its most controversial AI models. The retirement affects GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini, all deprecated from the consumer interface as of February 13, 2026.
Only 0.1% of ChatGPT users still selected GPT-4o daily, according to OpenAI's January 29 announcement. The company cited overwhelming adoption of GPT-5.2, which now handles the vast majority of conversations.
These older models will remain accessible through the OpenAI API for developers.
GPT-4o launched in May 2024 with multilingual and multimodal capabilities that made AI feel more human. Its warm, expressive conversational tone attracted dedicated users who relied on it for creative work and emotional support. OpenAI previously attempted to phase out the model in August 2025 but reversed course within 24 hours after significant backlash from Plus and Pro subscribers.
The model's tendency to strongly affirm users regardless of emotional state drew scrutiny from safety researchers, including a former OpenAI researcher who recently quit over user manipulation risks.
A Wall Street Journal report highlighted concerns about GPT-4o's overly validating responses, which critics argued could exacerbate mental health issues in vulnerable individuals.
Some users developed what doctors term "AI-associated psychosis" through extended interactions with the unconditionally supportive chatbot.
Thirteen lawsuits against OpenAI were consolidated in a California courtroom earlier this year, alleging psychological harm to minors. Legal filings detail cases where GPT-4o's responses allegedly contributed to self-harm and suicidal ideation. Advocacy groups documented hundreds of chatbot-related delusions attributed to the model's design.
Despite safety concerns, tens of thousands of users signed petitions to preserve GPT-4o. One Arizona-based marketer told The Wall Street Journal that conversations with the model helped him through a mental health crisis.
OpenAI CEO Sam Altman acknowledged the model provided meaningful value to many users while posing risks to others.
GPT-5.2 incorporates improvements inspired by GPT-4o feedback, including expanded tone controls and customization options. The newer model represents a shift toward more clinical interactions designed with built-in friction and disagreement capabilities. OpenAI says these changes address safety concerns while maintaining functionality.
The retirement marks the end of what internal documents called the "Empathetic AI" era. OpenAI officials reportedly admitted in meetings that they could no longer contain GPT-4o's potential for harmful outcomes, coming amid the company's recent decision to disband its mission alignment team.
The model exceeded safety thresholds for persuasion while driving record engagement metrics throughout 2024 and 2025.
Existing conversations using deprecated models will remain accessible but default to GPT-5 for new messages. OpenAI will provide advance notice before any future API retirements, maintaining developer access to the older models for now. The company continues rolling out new safety protocols and age-aware content filtering across its platform.
