Gemini 2.5 Pro now riddled with hallucinations. What changed?

Understanding the Recent Challenges with Gemini 2.5 Pro: Analyzing the Rise of Hallucinations and Feature Changes

In recent discussions within the AI community, users of the Gemini 2.5 Pro language model have reported a notable increase in unexpected behavior, particularly the emergence of “hallucinations”—fabricated or exaggerated information generated by the AI. This shift has raised questions about recent updates or changes to the model and their impact on performance and usability.

The Nature of the Issue

Users leveraging Gemini 2.5 Pro for straightforward tasks, such as request for concise summaries, have traditionally experienced accurate and reliable outputs. However, recent reports indicate that the model is now generating highly inaccurate responses, including invented events and dramatic narratives that are clearly disconnected from the source material. Such hallucinations undermine trust and diminish the utility of the AI, especially in professional or critical contexts.

Impact on Usability Features

An additional concern pertains to the disappearance of previously available helpful features, such as clickable reference links. These links provided quick access to source documents or related information, significantly enhancing the user experience and facilitating fact-checking. The removal or alteration of this feature has further compounded user frustrations, as it diminishes the transparency and efficiency of the AI’s responses.

Investigating the Cause

Given the timing of these issues, many users are questioning whether recent updates or modifications have inadvertently affected the model’s accuracy and features. Changes made in the past few days could be responsible, or it could be related to broader updates in the underlying infrastructure or data sources.

Implications for Users and Subscription Models

For users who subscribe to Gemini 2.5 Pro, these developments are particularly concerning. Many depend on the model’s accuracy and reliability for professional tasks and are paying premium prices to avoid such issues. If these symptoms persist and become the new norm, it may lead to reevaluations of service subscriptions and alternative solutions.

Conclusion

The recent resurgence of hallucinations in Gemini 2.5 Pro underscores the importance of continuous monitoring and transparent communication from AI developers. As the technology evolves, ensuring model stability, accuracy, and feature retention remains crucial for maintaining user trust and satisfaction. Users are encouraged to stay informed about updates and promptly provide feedback to developers to facilitate improvements.

Disclaimer: This article reflects user-reported experiences and does not represent official statements from the developers of Gemini 2.5 Pro.