Following the recent launchGPT-5.5Subsequently, OpenAI announced a complete upgrade of ChatGPT's default underlying model to..."GPT-5.5 Instant"This update not only significantly reduced "hallucinations" by more than half in internal testing, but also made significant adjustments to the response style—becoming more direct, reducing unnecessary follow-up questions, and toning down the use of emojis. Furthermore, to reassure users about the AI's personalized capabilities, OpenAI simultaneously launched a new "Memory sources" feature, allowing users to clearly understand which past conversations or data the AI used to generate personalized responses.
More precise, more efficient: 52.5% reduction in hallucinations.
According to data released by OpenAI, the upgraded GPT-5.5 Instant has made significant progress in improving the issue of "serious nonsense." In internal testing, the new model reduced hallucinatory claims by 52.5%; and in dealing with conversational situations that were previously flagged by users as "factually incorrect," inaccurate claims were reduced by as much as 37.3%.
In addition to improved accuracy, GPT-5.5 Instant's "personality" has also been tweaked. OpenAI states that the new model aims to provide more direct responses while retaining ChatGPT's original "warmth and personality." For users who frequently use ChatGPT as a productivity tool, the most noticeable change is that the new model will reduce "unnecessary follow-up questions" and, through more restrained use of emojis, make the output layout cleaner and more concise.
Personalization is further upgraded, returning "control of memory" to the user.
To address the pain point of users having to repeatedly enter background information, OpenAI has accelerated the speed at which its new model searches uploaded files, linked Gmail accounts, and past ChatGPT conversations to generate highly personalized responses.
However, high personalization comes with privacy concerns. To address this, OpenAI launched the "Memory Source" feature, allowing users to clearly see which data or conversation content the AI used to generate its responses.
• Transparent context tracking:This feature will allow users to clearly see which past information ChatGPT uses to customize their current response.
• Complete data autonomy:Users can choose to delete conversations that are no longer needed, or directly modify saved memories in the settings. If they only want to have a one-time conversation, they can also use the "Temporary chats" mode, which will not update the memory.
• Privacy protection when sharing:OpenAI emphasizes that when users choose to share conversation links with others, these "source of memory" markers will not be displayed.
Launch Schedule
• GPT-5.5 Instant model:Starting today, the rollout will begin to cover all users.
• Enhanced personalization features:It will be available first to Plus and Pro subscribers, while free, Go, Business, and Enterprise subscribers will gradually gain access in the coming weeks.
• Memory source function:It is expected to be fully available to all ChatGPT users on the web version and mobile app "in the near future".
