As expected, as GPT-5.3-Codex After the debut,OpenAI and has also been successively releasing GPT-5.3 and other models, with the one that debuted earlier being GPT-5.3 InstantThis is ChatGPT the most commonly used model. According to officials, this update focuses on making everyday conversations smoother and more practical, addressing issues that many users have long complained about, such as responses being too cautious or interrupting conversations. In short, it’s fixing the overly cautious behavior and becoming a “partner that gives direct advice.”
GPT-5.3 Instant Arrives: More Natural Conversations, More Direct Responses
GPT-5.3 Instant is a new instant chat model in the GPT-5 series. Compared to its predecessor, GPT-5.2 Instant, the key improvement is that responses are more direct, practical, and natural.
The older GPT-5.2 Instant often starts with lengthy explanations about its limitations, which makes conversations feel awkward or forces them to be interrupted. GPT-5.3 Instant significantly reduces this issue and directly provides helpful content. For example, when it comes to real-time information or sensitive topics, the new version doesn’t overemphasize “I can’t provide that” but instead tries to give answers that can be used as reference. Additionally, the new version has more mature web search integration and can provide additional context rather than just listing links. Writing is also more mature—GPT-5.2 Instant’s poetry tends toward abstraction, while GPT-5.3 Instant is more concrete and relatable.
In terms of performance testing, the data released by OpenAI this time focuses on reducing “hallucination rates,” including a 26.8% reduction when using web search compared to the previous generation, and a 19.7% reduction when relying solely on internal knowledge. In general user feedback evaluations, the hallucination rate for answers using the web decreased by 22.5%, and by 9.6% when not using the web. These numbers show that GPT-5.3 Instant has significantly improved in factual accuracy and is more reliable to use, especially in scenarios requiring precise answers.
Currently, all ChatGPT users can directly experience GPT-5.3 Instant, whether on free or paid accounts. Developers can access it using the name “gpt-5.3-chat-latest” in the OpenAI API.
For those who still prefer the old version of GPT-5.2 Instant, if you’re a paid subscriber, you can still find it in the “Current Models” menu. It’s expected to remain available for three months, until June 3, 2026, when it will be discontinued. Advanced versions like Thinking and Pro will be released at a later date.
Key features of GPT-5.3 Instant are as follows:
- More precise judgment of when to decline requests, reducing excessive unnecessary refusals and defensive behaviors from the previous model version.
- Deliver more accurate and contextually complete answers when searching the web is needed, while balancing online information with the model’s own knowledge to provide more valuable responses.
- Use a more natural style, get straight to the point, and avoid declarative statements.
- Further reduce hallucination rates when generating content across various topics
- Enhanced writing abilities to assist in creating highly imaginative and immersive content
Source: KOCPC Chinese