anyone else noticed that chatgpt 4o activates thinking quite often in the last few days? is there reasoning in 4o under the hood now?
In some cases, GPT-4o begins to apply reasoning before searching. This could be an initial indication of further updates and preparations for GPT-5 (tests?), where GPT-5 is known to be a one-in-all model. https://t.co/y5MQgSrOAB
all the non reasoning models in chatgpt now "think" when they search besides 4.1-mini. i wonder if theyre using the new small open source model openai is releasing to search 👀 https://t.co/WPPN25alar
OpenAI's GPT-4.5, released in February 2025, represents a major advancement in unsupervised scaling and pattern recognition, reducing hallucinations by approximately 40% compared to GPT-4o. Recent observations indicate that GPT-4o has begun to incorporate reasoning processes prior to conducting searches, a feature previously absent or limited. This new behavior, described as "thinking first before searching," appears across most non-reasoning ChatGPT models except for 4.1-mini. These developments suggest ongoing silent updates and potential preparations for the upcoming GPT-5 model, which is anticipated to be an all-encompassing AI system. Additionally, custom ChatGPT connectors using the Model Context Protocol (MCP) currently require both "search" and "fetch" tools to comply with OpenAI's technical specifications and are primarily utilized for deep research purposes.