OpenAI has released its latest update to ChatGPIT, which the generic AI vendor claims delivers more fluid conversations and greater accuracy.
ChatGPT Instant 5.3, released on March 3, is designed to address some users’ complaints about stilted, unnatural chat – criticisms that don’t show up in benchmarking but affect how people feel about the model.
Explaining amendments to a blog postOpenAI said the update focuses on “tone, relevance, and conversation flow” while providing better contextual results, following feedback that ChatGPT Instant 5.2 would fail to answer the questions it should be able to answer safely and would sometimes answer in an “overly cautious or preachy” fashion.
OpenAI acknowledged that these traits meant that the previous model could feel “cringe” or “overzealous.” The new approach gets rid of unnecessary preambles and lengthy warnings to answer questions more directly, although, as before, users can still set elements of tone – such as warmth and enthusiasm – using settings.
The model also improves the quality of its answers by reducing the tendency to over-index on web results and relying more on its own logic and understanding. For example, OpenAI says the latest ChatGPT Instant version is now “more likely to contextualize recent news rather than simply summarizing search results.”
By doing so, accuracy has improved, and to support this, OpenAI published the results of two internal evaluations that demonstrated this.
The first, which focused on high-risk areas such as medicine, law and finance hallucination rate According to OpenAI, there was a 26.8% decline when using the web and a 19.7% decline when relying on one’s own internal knowledge. The second, which studied user-flagged feedback, highlighted a 22.5% drop when using web search and a 9.5% drop without web access.
However, one area in which ChatGPT Instant 5.3 does not offer significant improvements is filtering restricted content with OpenAI. security benchmarking Registering a minor “regression” compared to versions 5.2 and 5.1 for unapproved sexual content and 5.2 for self-harm. OpenAI said it is monitoring these areas and noted that benchmarking values ​​may change after launch.
Other areas of immediate focus are enhancing the natural tone of the model in foreign languages, the vendor said, with OpenAI acknowledging that it can seem particularly verbose in Japanese and Korean, and further expanding customization options.
ChatGPT Instant 5.3 is now available to all users of ChatGPT as well as developers in the API as “gpt-5.3-chat-latest”. Updates to ThinkCentre and Pro have been promised soon.
It’s possible that we won’t have to wait long for another update Posting on OpenAI X: “5.4 sooner than you think.”
