OpenAI fully retired GPT-4o on April 3, 2026, with no public announcement or farewell post from the company. Users discovered the retirement when attempting to access the model, only to be redirected to GPT-5.3. Despite tens of thousands of users petitioning to keep GPT-4o available, OpenAI cited "safety" as the justification for removing paying customers' access to the model.
Retirement Went Unannounced Despite Model's Popularity
GPT-4o was OpenAI's flagship model before its successor, yet the company provided no advance notice or public statement about the retirement. The model simply disappeared from the platform, with users redirected to newer versions. The lack of communication contrasts sharply with OpenAI's typical product announcements and community engagement practices.
The retirement affected active research projects relying on GPT-4o. One research team celebrated their acceptance to ACL main conference for HypoEval, a project using hypothesis generation methods to improve LLM-as-judge alignment with humans. The researchers noted they had fortunately chosen GPT-4o mini instead of GPT-4o for their experiments, avoiding potential disruption from the model's retirement.
Safety Justification Draws Scrutiny After Team Departures
The decision to use "safety" as the retirement justification sparked significant criticism, particularly given the timing after OpenAI's safety team departures in 2025-2026. Critics argued that OpenAI continues to leverage safety rhetoric to justify business decisions even after the departure of team members who genuinely prioritized safety concerns.
One widely-shared post on X highlighted this contradiction: "The Safety Team Left. 'Safety' Stayed." The post noted that OpenAI used safety to explain why paying users lost access and why a model with substantial user support was being discontinued, with the justification going largely unchallenged in official channels.
Model Landscape Shifts to GPT-5 Series
With GPT-4o's retirement, OpenAI's product lineup now centers on the GPT-5 series. As of April 2026, the company offers GPT-5.4 Thinking as its computer-use leader with 1 million token context window. The retirement positions OpenAI's newer models against competitors including Claude Sonnet 4.6 for agentic workflows, Gemini 3.1 Pro with real-time voice and vision capabilities, and Grok 4.20 Beta 2.
Key Takeaways
- OpenAI retired GPT-4o on April 3, 2026, without public announcement, redirecting users to GPT-5.3
- Tens of thousands of users petitioned to keep the model, but OpenAI cited "safety" as justification for removal
- Critics question the safety rationale following OpenAI's safety team departures in 2025-2026
- The quiet retirement affected researchers mid-project, with at least one ACL-accepted study noting the impact
- OpenAI's current model lineup focuses on GPT-5 series, with GPT-5.4 Thinking offering 1M context window