OpenAI has announced the launch of a new model, ChatGPT 5.3 Instant, which shall now appear by default for all ChatGPT users and for developers using the API under the name “gpt-5.3-chat-latest.” In turn, GPT‑5.2 Instant will remain available for three months to users on paid plans until it is retired on June 3, 2026.
But… why has OpenAI decided to implement this change?
Let us see what OpenAI itself states in its official announcement:
“The tone of GPT‑5.2 Instant at times felt uncomfortable because it could be authoritarian or make unwarranted assumptions about the user’s intentions or emotions. This update provides a conversation style that is more focused and natural, with fewer proclamations and unnecessary phrases such as ‘Stop. Take a deep breath.’ We are also seeking to maintain ChatGPT’s personality consistently across conversations and updates, so that improvements are experienced as increased capabilities without sacrificing a familiar and stable experience.”
It is interesting, because OpenAI uses the term “authoritarian” to refer to GPT 5.2, when it might be more appropriate to qualify it as “condescending.” Very, very condescending.
In fact, this is not even a personal opinion. Many users around the world have expressed their disagreement with ChatGPT’s style of responding to their queries, which often resulted in what felt more like a virtual counseling session than a serious and direct answer to a specific question. It is evident that, on some occasions, opting for such a compassionate tone can be useful for certain users; however, for others, it is even potentially dangerous, as it may overlook certain warning signs related to the safety or health of those making the inquiries.
Difference in the approach to answering in ChatGPT 5.3, less condescending
If we consult the official statement, OpenAI itself provides hints regarding user complaints, even stating that “this update focuses on those aspects of ChatGPT that people experience every day: tone, relevance, and conversational fluency. These are subtle details that do not always show up in benchmark tests, but that make the difference between a helpful experience and a frustrating one.”
As explained in its introductory tweet, “greater accuracy, less cringe.”
The 5.3 Instant version introduces improvements that directly impact the daily workflow, both for casual users and for developers who rely on the API. Overall, OpenAI seeks for GPT‑5.3 Instant to provide more accurate answers, richer results, and better contextualization when searching the web, reducing warnings and overly declarative language. Moreover, GPT‑5.3 Instant reduces hallucination rates by 26.8% when using the web and by 19.7% when relying exclusively on its internal knowledge, compared to previous models.
As noted previously, one source of frustration with version 5.2 was its tendency to deliver moral lessons before responding to a technical or creative question. OpenAI has recalibrated the alignment system so that the model is more pragmatic, laconic, and direct. Now, if you request complex code or an objective data analysis, the model understands that you are seeking efficiency and not emotional or psychological validation.
Example response from the new ChatGPT 5.3. Source: OpenAI
OpenAI acknowledges that, by softening the authoritarian tone, the model is now marginally more likely to accept requests it previously blocked out of excessive caution. While this represents a victory for creative freedom, it also presents a challenge for moderation. The new approach is based on intervening only when it is strictly necessary.
However, not everything is related to conversational tone. True to its Instant moniker, this model has optimized the Time To First Token. The model’s architecture allows for text generation speeds approaching 1,200 tokens per second, eliminating that sense of pause for contemplation.
Previously, when ChatGPT searched the web, it could lose track of the original conversation or limit itself to presenting a list of links. With version 5.3, external information is integrated organically into responses, and sources are provided in a manner that is more subtle yet precise, so that the text no longer resembles a bibliography.
Additionally, GPT‑5.3 Instant is less prone to focusing excessively on web results, utilizing its prior knowledge to contextualize recent news instead of simply summarizing search results.
According to OpenAI, one recurring complaint from developers was that the model would lose its tone midway through an extended conversation. OpenAI has introduced an improvement within the context window that prioritizes the original system instructions, ensuring that if you ask for responses to be “concise and technical,” it will not revert to being “poetic and compassionate” after ten exchanges.
Image: Gemini
Your email address will not be published. Required fields are marked *
Δ