ChatGPT just got a conversational upgrade.
OpenAI has rolled out GPT-5.3 Instant, a new version of its most widely used model designed to deliver smoother dialogue, more accurate responses, and stronger answers when pulling information from the web.
The tech giant said the update focuses on improving the everyday experience of using ChatGPT, refining how the assistant responds so conversations feel more natural and consistently helpful.
“This update focuses on the parts of the ChatGPT experience people feel every day: tone, relevance, and conversational flow,” OpenAI said in a blog post. “These are nuanced problems that don’t always show up in benchmarks, but shape whether ChatGPT feels helpful or frustrating.”
Fewer dead ends in everyday conversations
GPT-5.3 Instant is built to answer more questions directly instead of shutting conversations down. The update reduces unnecessary refusals that sometimes appeared in GPT-5.2 Instant, where the system declined prompts it could safely address.
OpenAI also trimmed back overly cautious disclaimers and moralizing preambles that could slow responses. The AI model reaches the answer faster, a change that reflects user feedback about frustrating conversational dead ends that interrupted otherwise simple interactions.
From search results to synthesized answers
The new model also improves how ChatGPT handles questions that rely on web-based information. GPT-5.3 Instant now does a better job balancing online results with its own reasoning rather than simply summarizing what it finds.
The system can identify the key point of a query and surface the most relevant information first. The change is meant to make responses more useful when people ask about current events, recent developments, or other time-sensitive topics that depend on web data.
Lower hallucination rates
GPT-5.3 Instant brings stronger factual reliability across a range of topics. OpenAI evaluated the model on internal benchmarks tracking hallucinations, including tests covering higher-stakes areas such as medicine, law, and finance.
The results show clear reductions in incorrect or fabricated information. Hallucination rates dropped 26.8% when web data was used and 19.7% when relying on internal knowledge, compared with earlier models.
In a separate evaluation based on de-identified ChatGPT conversations flagged by users for factual errors, hallucinations fell 22.5% with web access and 9.6% without it.
Writing gets an upgrade
ChatGPT’s writing abilities also see a lift with the update, giving users greater range across both creative and practical tasks. Drafts can carry more texture and coherence, whether the goal is shaping a narrative passage, polishing a paragraph, or working through ideas on the page.
The update is rolling out now in the AI tool and is also available to developers through the API under the name gpt-5.3-chat-latest. The previous version, GPT-5.2 Instant, will remain accessible in the Legacy Models section of the model picker before its scheduled retirement on June 3, 2026.
Riley Walz, known for his viral Silicon Valley tech pranks, has joined OpenAI’s OAI Labs.