Recent user observations indicate that OpenAI's ChatGPT 4o model has started to exhibit a new behavior described as "thinking" before searching or responding. This change appears to be consistent across several versions, including GPT-4o and GPT-4.5, and is also present in most non-reasoning ChatGPT models except for the 4.1-mini variant. Some users speculate that this feature may be linked to a new small open-source model released by OpenAI to enhance search capabilities. While the "thinking" function suggests an underlying reasoning process, some users report that ChatGPT 4o's performance has deteriorated, citing increased hallucinations and poor mathematical accuracy, which has affected its usability.