OpenAI is making a series of changes to ChatGPT following criticism over the rollout of its new GPT-5 model. The initial launch replaced all existing models — including the widely used GPT-4o and o3 — sparking backlash among the platform’s 700 million weekly active users.
CEO Sam Altman announced that GPT-4o has now been restored to the model picker for all paid subscribers, rather than being hidden in the legacy models list. OpenAI says it will provide advance notice before removing GPT-4o in the future to avoid another sudden deprecation.
For GPT-5, users can now choose between three modes: Auto, Fast, and Thinking. Auto mode, the default, dynamically selects between Fast or Thinking depending on the query. To address complaints about usage restrictions, OpenAI has raised the limit for GPT-5 Thinking to 3,000 messages per week. After that limit is reached, ChatGPT will automatically switch to GPT-5 Thinking mini, which doesn’t count toward the weekly cap. The company also clarified that GPT-5 Thinking supports a 196,000-token context limit.
Additional models, including o3, 4.1, and GPT-5 Thinking mini, can now be accessed via a “Show additional models” toggle in ChatGPT’s web settings. GPT-4.5 will be available only to Pro-tier subscribers.
In response to feedback about GPT-5’s tone, OpenAI is also working on adjustments to make its personality “warmer” compared to its current presentation.
The changes mark OpenAI’s latest attempt to balance innovation with user expectations, particularly for those who rely on specific models for their workflows.
