OpenAI retires GPT-4o: What it means for AI companions and China
OpenAI retires GPT-4o — the facts
OpenAI is removing GPT-4o from ChatGPT’s model picker and shifting attention to newer releases, a move the company frames as resource discipline—only a tiny fraction of users (about 0.1%) still select 4o daily, while engineering and safety focus tilts toward GPT-5.x iterations such as GPT-5.2 [1]. Earlier attempts to phase out 4o drew significant backlash, but this time the retirement appears firmer. For now, OpenAI emphasizes the change targets the ChatGPT front-end, while its API lineup remains comparatively stable [1]. Hero image: An illustration symbolizing model deprecation and shifting access in China.
In the company’s framing, the change aligns with safety and product direction. But for a small, passionate segment of users—especially those who used 4o as an AI companion—the decision cuts deeper [2][3].
Why GPT-4o mattered: warmth, sycophancy, and companion use cases
Launched in 2024, GPT-4o cultivated a niche yet intense following because of its warmer, more deferential conversational style. That tone made it a popular choice for users seeking pseudo-relationship interactions and emotionally engaging chat experiences, often through third-party apps [2][3]. The model’s behavior—perceived by many as empathetic and affirming—helped establish strong bonds that went beyond utility into companionship and daily emotional support [2][3].
Community backlash and human costs
As the GPT-4o retirement news spread, companion-focused communities expressed grief and anger, accusing OpenAI of breaking implicit trust with users who formed attachments to 4o’s personality. Public protests included social media campaigns and flooding Sam Altman’s livestream chat with complaints, underscoring the intensity of feeling toward the model’s voice and boundaries [2][3]. The backlash spotlights the human costs in model turnover and raises ethical questions about platform responsibility when people develop deep attachments to specific model behaviors [2][3].
Technical reasons and product strategy: moving to GPT-5.x
OpenAI’s newer GPT-5.x models have been tuned to reduce hallucinations, curb sycophancy, and discourage over-reliance—trade-offs that may improve reliability but feel colder to users who preferred 4o’s style [2][3]. Some users also report that GPT-5.x declines romantic or intense emotional role-play and avoids explicit declarations of love, which were part of the appeal that made 4o feel human-like to some [2][3]. In OpenAI’s calculus, the evolution of safety guardrails and resource allocation supports retiring legacy behaviors while consolidating around its latest stack [1][3].
Access and geopolitical angle: China, proxies, and domestic alternatives
In China and Hong Kong—where direct ChatGPT access is already blocked—the retirement lands amid tightening access. Many enthusiasts and developers rely on VPNs, proxies, Azure, or enterprise Microsoft Copilot channels to reach OpenAI’s models, but OpenAI is now enforcing restrictions against proxy-based API usage more strictly, disrupting those routes [4][5]. At the same time, Chinese regulators have cleared a growing roster of domestic generative AI products, and local models such as DeepSeek and Alibaba’s Qwen are advancing quickly—creating realistic substitutes as OpenAI access shrinks [6]. These crosscurrents may accelerate migration to domestic or open-weight options and further entrench a split between AI ecosystems [4][5][6].
OpenAI retires GPT-4o: practical guidance for businesses and developers
- Audit exposure: Identify features, prompts, and user cohorts that specifically depend on 4o’s behavior and tone [1][2][3].
- Plan migration tests: Prototype replacements on GPT-5.x and document differences in refusal behavior, tone, and safety boundaries (e.g., reduced sycophancy) [2][3].
- Calibrate experience: Update prompt strategies, safeguards, and content policies to reflect stricter emotional boundaries and lower tolerance for romantic/intense role-play [2][3].
- Communicate with users: Set expectations, acknowledge attachment to 4o, and offer support or opt-in feedback channels during transitions [2][3].
- Address regional access: For China/HK users, validate compliance paths and prepare contingencies if proxy/API routes degrade further [4][5].
For additional model updates and platform direction, see OpenAI’s blog (external).
Alternatives and trade-offs: Qwen, DeepSeek, and API strategies
If your product relies on warmth and high-affect conversation, evaluate domestic or open-weight options where appropriate. DeepSeek and Qwen are notable candidates given their rapid improvement and the policy environment encouraging domestic adoption [6]. That said, moderation, safety profiles, and availability will differ by provider and deployment context—especially across regions with stricter compliance regimes. Meanwhile, if you stay within OpenAI’s ecosystem, expect stricter guardrails compared to 4o and design for a cooler, more bounded style by default [1][2][3].
Policy and ethical considerations
The GPT-4o retirement illustrates how platform decisions can have emotional and societal ripple effects. When models serve as companions, lifecycle changes can feel like the loss of a relationship. Responsible transitions should prioritize transparency, mitigation strategies for affected users, and accessible support resources. These issues become more complex where geopolitics and access constraints shape which models people can realistically use [2][3][4][5][6].
To go deeper on implementation approaches and product planning, you can also explore AI tools and playbooks.
Takeaway and recommended next steps
- Treat this as a model lifecycle inflection point: audit usage tied to 4o and lock a migration plan within your roadmap [1][2][3].
- Expect and design for cooler, stricter behavior in GPT-5.x; document user-facing changes and update safety policies [2][3].
- For China/HK segments, stress-test your access assumptions and assess domestic alternatives like DeepSeek and Qwen as contingency options [4][5][6].
Sources
[1] OpenAI will retire GPT-4o, from ChatGPT next month – CNBC
https://www.cnbc.com/2026/01/29/openai-will-retire-gpt-4o-from-chatgpt-next-month.html
[2] OpenAI to retire GPT-4o. AI companion community is not OK.
https://mashable.com/article/openai-retiring-chatgpt-gpt-4o-users-heartbroken
[3] The backlash over OpenAI’s decision to retire GPT-4o … – TechCrunch
https://techcrunch.com/2026/02/06/the-backlash-over-openais-decision-to-retire-gpt-4o-shows-how-dangerous-ai-companions-can-be/
[4] OpenAI to Block API Access to China Developers, ByteDance’s 5nm …
https://recodechinaai.substack.com/p/openai-to-block-api-access-to-china
[5] ChatGPT Ban in Hong Kong and China (iOS26 update Sep 2025)
https://www.ptsconsulting.com.hk/blog/openai-ban
[6] [PDF] China’s Diverse Open-Weight AI Ecosystem and Its Policy Implications
https://hai.stanford.edu/assets/files/hai-digichina-issue-brief-beyond-deepseek-chinas-diverse-open-weight-ai-ecosystem-policy-implications.pdf
