OpenAI has brought back the model picker in ChatGPT following weeks of backlash over its removal during the GPT-5 rollout. The change marks a major shift in approach, giving users back the ability to choose between different models rather than relying solely on automated selection.
When GPT-5 was introduced, OpenAI aimed to streamline the user experience by automatically deciding which model handled a query. While the intention was to reduce complexity, it quickly became clear that the approach alienated many users. Developers, researchers, and even casual users reported that automatic switching often delivered inconsistent results, and some noted a loss of preferred model characteristics that were essential to their workflows.
The restored model picker is available to paid ChatGPT subscribers, with GPT-4o now offered as the default model for many. Users can enable a “show additional models” option to access o3, GPT-4.1, and GPT-5 Thinking mini. Pro-tier subscribers can also continue using GPT-4.5, a model known for its high accuracy in technical and research-heavy tasks. The return of these options means users once again have control over speed, style, and performance depending on their needs.
OpenAI has paired the model picker’s return with the launch of new “speed modes” that let users fine-tune response behavior. The Auto mode adapts to the query’s complexity, Fast mode prioritizes speed for shorter, more straightforward prompts, and Thinking mode gives the model extra time to reason through complex problems. This added flexibility directly addresses one of the key criticisms of the earlier change, where users felt their experience was being overly simplified at the expense of precision.
Another major update is OpenAI’s commitment to transparency. The company has promised to provide advance notice before retiring any models, avoiding the sudden removal of tools that users rely on daily. This pledge comes after the frustration sparked when GPT-4o was initially phased out without warning, forcing some users to rework established processes overnight.
Alongside the restored model picker, GPT-5 itself has been updated. The model’s personality has been refined to feel warmer and more natural in conversation, countering early feedback that it was overly formal or detached. Usage limits have also been expanded, allowing subscribers to run more queries before hitting daily caps. For users who prefer GPT-5’s advanced reasoning, these changes make it a more appealing option without locking out those who rely on older models for specific tasks.
