Users of ChatGPT recently noticed a new “alpha models” section in the model selection, where experimental agents briefly appeared with names like “Agent with truncation” and “Agent with prompt expansion.” These models activated agent mode, meaning ChatGPT automatically used a browser or additional tools to complete the user’s task.
The new agents appeared only for a short time and quickly disappeared from the program, indicating an accidental feature launch on the server. Differences in names, such as “truncation” and “prompt expansion,” suggest tests of different system prompt settings or even new model architectures. This could be preparation for the launch of agents based on GPT-5, as previous versions of agents operated on older models.
Features of this type will primarily interest researchers, enthusiasts, and users who actively test AI’s work with tools. They can provide more control over how AI processes long queries or expands prompts for complex tasks, such as during information searches or analyzing large volumes of data.
OpenAI has not yet provided official explanations regarding the new agents, but CEO Sam Altman previously announced significant updates and the launch of new features soon. Such tests usually occur in the final stages of development before a broader launch and may change the way of working with ChatGPT for those who use AI for research or automation of everyday tasks.
