OpenAI starts rolling out GPT-5.3 Instant in ChatGPT as default everyday model

OpenAI says the update targets fewer unnecessary refusals and a smoother, more direct conversational flow, while keeping the same safety approach as prior releases.

OpenAI has begun rolling out GPT-5.3 Instant in ChatGPT, introducing changes designed to make everyday conversations more direct, relevant, and fluid.

The update targets issues users reported with the previous GPT-5.2 Instant model, including overly cautious responses, excessive disclaimers, and answers that could feel slow to reach the point.

The model update is becoming available to ChatGPT users and developers through the API, where it appears under the name gpt-5.3-chat-latest. OpenAI says GPT-5.2 Instant will remain available to paid users in the model picker under Legacy Models for three months before being retired on June 3, 2026.

The release reflects a broader focus on refining everyday AI interactions rather than introducing a new architecture. OpenAI says improvements target tone, relevance, and conversational flow, areas that shape how people use ChatGPT in tasks such as research, writing, and learning.

OpenAI responds to user feedback on ChatGPT responses

Nick Turley, Head of ChatGPT at OpenAI, announced the rollout in a LinkedIn post, outlining the feedback that informed the update. : “GPT-5.3 Instant is rolling out in ChatGPT starting today.

“One theme we kept hearing from users about 5.2 was that responses could sometimes be too cautious, overloaded with caveats, or just not flow as naturally as they should in conversation. Those are subtle issues, but we know they really shape whether ChatGPT feels helpful or frustrating to use day to day.

“GPT-5.3 Instant focuses on that feedback with fewer unnecessary refusals, fewer defensive disclaimers, and answers that get to the point more directly while still being thoughtful and safe.

“Really proud of the team on this one and continuing to ship improvements that people feel in everyday conversations.”

According to OpenAI, the update reflects recurring feedback from ChatGPT users about how models respond in real conversations. While benchmarks measure technical performance, the company says everyday usability often depends on subtler factors such as tone and conversational pacing.

Fewer refusals and more direct answers

One of the main changes in GPT-5.3 Instant relates to how the model handles requests that fall close to safety boundaries.

OpenAI says previous versions could sometimes refuse questions that were safe to answer or lead with lengthy explanations about what the system could not do. GPT-5.3 Instant reduces unnecessary refusals and removes overly defensive preambles, allowing the model to respond more directly when an appropriate answer is possible.

The company says the goal is to reduce “dead ends” in conversations while maintaining existing safety guardrails.

Improved web answers and lower hallucination rates

The update also improves how ChatGPT handles questions that involve information from the web.

OpenAI says GPT-5.3 Instant better balances online sources with its own knowledge and reasoning. Earlier models could over-rely on search results, producing long lists of links or loosely connected information. The updated system is designed to recognize the intent behind questions and surface the most relevant information earlier in a response.

OpenAI also reports lower hallucination rates in internal testing.

On a higher-stakes evaluation covering fields such as medicine, law, and finance, hallucinations dropped by 26.8 percent when using the web and 19.7 percent when relying on internal knowledge compared with earlier models.

In a separate evaluation based on real ChatGPT conversations flagged by users for factual errors, hallucinations decreased by 22.5 percent with web use and 9.6 percent without web access.

Availability and next steps

GPT-5.3 Instant is now rolling out to ChatGPT users and is available to developers through the OpenAI API. OpenAI says GPT-5.2 Instant will remain accessible in the model picker under Legacy Models for paid users during a three-month transition period. After that, the model will be retired.

The company says updates to other ChatGPT model tiers, including Thinking and Pro models, are expected to follow.

For the EdTech sector, the release highlights how AI model improvements are increasingly focused on usability rather than only benchmark performance. As conversational AI tools become embedded in classrooms, research workflows, and learning platforms, factors such as tone, clarity, and contextual reasoning are becoming more relevant for both students and educators.

ETIH Innovation Awards 2026

The ETIH Innovation Awards 2026 are now open and recognize education technology organizations delivering measurable impact across K–12, higher education, and lifelong learning. The awards are open to entries from the UK, the Americas, and internationally, with submissions assessed on evidence of outcomes and real-world application.