GPT-5 too unfriendly: OpenAI reverts to 4o as standard model
After one week of GPT-5, OpenAI responds to criticism: paying users get 4o back as standard. GPT-5 routing now also works in manual mode.
(Image: Tada Images/Shutterstock.com)
GPT-5 does not strike the right note from the perspective of some users. The major update to OpenAI's Large Language Model (LLM) is not friendly enough for some users, according to forum posts. This and other issues have now apparently prompted OpenAI to switch the standard model for paying customers back to GPT-4o for now. OpenAI boss Sam Altman announced the change at X. It can also be found in the release notes.
Along with this change, users of GPT-5 can now also switch manually between the "Auto", "Fast" and "Thinking" modes. This is probably also a response to criticism because the automatic switch did not work properly for some users. This should actually make it easier to use the chatbot if users have to deal with fewer setting options. And OpenAI has introduced new connectors that allow users to connect the AI to services such as Google Docs – but not in Europe at this time.
Access to older models
"Most users will choose Auto, but the additional choice will be useful for some people," said Altman, explaining the change to the routing of GPT-5. In this context, he also revealed limitations of the model presented a week ago. Thinking mode, for example, is currently limited to 3000 messages per week. Anyone who exceeds this limit will receive additional messages via GPT-5 Thinking mini. The context limit is 196,000 tokens. According to Altmann, OpenAI will have to adjust the limits again depending on usage.
Videos by heise
In addition to 4o, paying users now also have quick access to other models via the web interface. Among others, o3, 4.1 and GPT-5 Thinking mini can be found there. However, in order for these to appear in the menu, we first had to activate "Show additional models" in the settings during testing. However, 4.5 is only available for Pro users who pay the more expensive tariff, as it requires a lot of hardware performance.
(Image:Â ho)
Personality is being revised
In connection with the criticism of the "personality" of GPT-5, Altman explained that OpenAI is working on an update that will make the new model appear "warmer". However, it should not be as "annoying" as GPT-4o. One lesson that OpenAI has learned from the criticism is that the personality of the model should be made more customizable by the user in the future, explained the OpenAI boss.
OpenAI also announced new connectors for ChatGPT. However, these are not available in Europe for now. Plus users can connect the services Canva, Dropbox, HubSpot, Notion, Microsoft SharePoint and Microsoft Teams to the AI. Pro users can also connect GitHub. The so-called connectors are still in the beta phase and can be activated via the settings. They allow the AI to access the user data for the respective services, eliminating the need for manual transfers. Connectors for Gmail, Google Calendar and Google Contacts are to follow.
(mki)
