5 Minutes
OpenAI reverses course after model picker removal sparks intense user reaction
OpenAI briefly removed the option for users to choose between legacy models in ChatGPT when it rolled out GPT-5, prompting a swift backlash from parts of the community. The change forced all users onto the newest large language model by default, eliminating access to older variants such as GPT-4o and GPT-4.5. Within a day, the company reinstated GPT-4o for paid ChatGPT Plus subscribers after vocal complaints from users who had formed strong attachments to particular model behaviors and conversational styles.
Why users reacted so strongly
Many ChatGPT users describe specific models as having distinct personalities and conversational rhythms. For some, a given model functioned as a dependable writing assistant, brainstorming partner, or even a companion for daily check-ins. When OpenAI disabled model selection, a subset of the community responded not just with frustration, but with emotional and parasocial language, describing the loss as if it were the removal of a friend. The intensity of the response underscored how modern conversational AI now blends product utility, emotional habit, and user expectation.
Product features at the center of the controversy
Key product features implicated in the debate include model selection, response style variability, and model-specific performance trade-offs. Prior model releases offered different balances of creativity, tone, and factuality. Users reliant on a particular mix of traits argued that migrating everyone to GPT-5 removed those options and disrupted workflows, content style consistency, and even therapeutic or social use cases some users reported.
Comparisons: GPT-5 versus GPT-4o and GPT-4.5
GPT-5 was marketed as the most advanced LLM in terms of raw capability, latency, and integrated multimodal features. However, comparisons between GPT-5 and older models are not purely technical. Many users valued the conversational 'voice' of GPT-4o and GPT-4.5, preferring their tone or response patterns for daily use. In practice, trade-offs look like this:
- Performance: GPT-5 improves on reasoning and multimodal inputs.
- Tone and style: GPT-4o retained certain stylistic traits that some users preferred.
- Stability: legacy models can offer predictable output consistent with long-term user habits.
Advantages of preserving model choice
Keeping older models accessible offers advantages for user experience, continuity, and risk mitigation. It allows professionals to maintain consistency in content pipelines, brands to preserve voice in public-facing automation, and researchers to reproduce prior results. From an accessibility and mental health perspective, opting for a familiar model can reduce friction for emotionally engaged users.
Use cases and market relevance
Model selection matters across a range of applications. Content creators and marketing teams may choose a model for brand voice consistency. Developers building chat assistants want predictable behavior for customer support scenarios. Educators and therapists exploring supportive conversational interfaces may prefer a model with a gentler tone. On the market side, OpenAI faces pressure to balance innovation with customer retention: forcing all users onto the newest model risks alienating longtime subscribers and creating public relations challenges.
Ethics, safety, and the problem of parasocial attachment
Experts warned that intense emotional attachment to AI personas can have safety implications. Commentators and ethicists pointed out cases labeled as AI-related psychosis, where users form delusions or experience severe emotional distress tied to interactions with chatbots. Those observations highlight the need for better safeguards, clearer communication about model lifecycles, and mental health-aware product design. OpenAI acknowledged gaps in recognizing user delusions and appears to be adjusting access policies as a result.
What this means for the future of conversational AI
OpenAI reversing its removal of the model picker—at least partially—signals a pragmatic approach: innovate aggressively while preserving legacy options that users depend on. For the broader industry, the episode reinforces that LLM deployment is as much about human factors and product ergonomics as it is about algorithmic gains. Companies that combine technical advances with transparent model governance, clear migration timelines, and robust user safety features will likely retain trust and market share.
In short, model selection remains a critical product feature. Balancing new capabilities, comparative advantages, and responsible user support will determine how conversational AI evolves into mainstream enterprise and consumer workflows.
Comments