With GPT-5's launch, OpenAI has removed its older models like GPT-4o and o3 from the ChatGPT model selector, sparking a backlash from some users
‘They have completely ruined ChatGPT,’ one user complains. Some are even canceling their paid subscriptions to ChatGPT, claiming GPT-5 is inferior to the company's previous models.
Context & Ripple Effects
OpenAI had already had to roll back a GPT-4o update over sycophantic behavior, showing that model changes can disrupt users’ established expectations. Removing familiar choices at the GPT-5 transition turns a quality dispute into an access-policy dispute.
The backlash also sits within a quickly evolving access policy: related coverage says OpenAI restored GPT-4o for paid users and promised advance notice after initially narrowing selection. That makes the selector itself, not only the newest model’s capabilities, a retention lever.
First-order effects
- ChatGPT users who relied on GPT-4o or o3 lose direct access to those workflows in the selector, while OpenAI concentrates their usage on GPT-5.
- Paid users voicing dissatisfaction have a clearer reason to reassess subscriptions: the service has changed both the available model and their ability to choose among alternatives.
Second-order effects
- OpenAI faces pressure to preserve legacy-model access, segment it by plan, or provide clearer deprecation notices; related coverage points to temporary old-model access for Pro users before a planned phaseout in the Pro-user transition plan.
- Competitors can position persistent model choice and stable workflow behavior as differentiation for users whose prompts or habits perform differently across models.
Third-order effects
- If providers increasingly replace models rather than maintain broad user choice, model retirement becomes a core subscription-governance issue, with customers treating continuity as part of the product they pay for.
- The episode strengthens model buyer power: even where a provider controls the endpoint, concentrated user cancellation or churn risk can constrain how abruptly it withdraws familiar models.
The trend: Frontier AI products are shifting from simple model launches toward active governance of legacy access, user choice, and workflow continuity.