Tech News

OpenAI Rolls Out GPT-5.3 Instant for ChatGPT with Fewer Hallucinations and Smoother Conversations

Highlights

  • OpenAI launched GPT-5.3 Instant for ChatGPT, improving tone, reducing unnecessary refusals, disclaimers and overly cautious phrasing.
  • The update lowers hallucination rate, offer smoother and more factual everyday conversations.
  • GPT-5.3 Instant is now live for all ChatGPT users and developers via API.

OpenAI Rolls Out GPT-5.3 Instant. (Image credit – OpenAI)

OpenAI has announced an update to its most widely used ChatGPT model with the launch of GPT-5.3 Instant. The new version is designed to deliver more accurate responses, improved contextual understanding during web searches and a more natural conversational tone.

According to the company, GPT-5.3 Instant reduces unnecessary digressions, excessive disclaimers and overly declarative phrasing. It also lowers hallucination rates, making responses more reliable across different use cases.

In a blog post titled “PT-5.3 Instant: Smoother, more useful everyday conversations,” OpenAI stated, “This update focuses on the parts of the ChatGPT experience people feel every day: tone, relevance, and conversational flow. These are nuanced problems that don’t always show up in benchmarks, but shape whether ChatGPT feels helpful or frustrating. GPT-5.3 Instant directly reflects user feedback in these areas.”

Improved Tone and Reduced Over-Cautious Responses

The previous GPT-5.2 Instant model was criticised for having what users described as a “cringe” tone. It could sometimes come across as overbearing or make assumptions about a user’s emotions or intent. GPT-5.3 Instant aims to address this by adopting a more natural and balanced conversational style, eliminating dramatic phrases such as “Stop. Take a breath.”

OpenAI also noted that GPT-5.2 Instant occasionally refused to answer questions it was capable of addressing, or responded with overly cautious language when handling sensitive topics. GPT-5.3 Instant reduces unnecessary refusals and tones down defensive or moralising introductions.

The model is designed to avoid “over-caveating” after assuming negative intent from users.

Better Web Integration and Fewer Hallucinations

GPT-5.3 Instant offers enhanced web-based responses by better balancing online information with its internal knowledge base. This improvement reduces the likelihood of over-reliance on web results.

OpenAI claims that the updated model provides more factual answers with significantly fewer hallucinations. In higher-stakes evaluations, GPT-5.3 Instant reduced hallucination rates by 26.8 percent when using web access and 19.7 percent when relying solely on internal knowledge, compared to earlier versions.

In user-feedback evaluations, hallucinations decreased by 22.5 percent with web usage and 9.6 percent without web access.

Limitations and Ongoing Improvements

While GPT-5.3 Instant represents a notable step forward in everyday usability, OpenAI acknowledged that further refinements are still needed.

For non-English languages such as Japanese and Korean, responses can sometimes sound overly literal or unnatural. Improving tone and conversational flow across multiple languages remains a priority for the company.

Additionally, although GPT-5.3 Instant introduces a smoother response style, OpenAI says it will continue monitoring user feedback and enhancing customisation options over time.

Availability Details

GPT-5.3 Instant is now available to all users in ChatGPT. Developers can also access the model via the API under the name ‘gpt-5.3-chat-latest.’ Updates to Thinking and Pro tiers are expected to roll out soon.

Meanwhile, GPT-5.2 Instant will remain accessible for three months for paid users under the Legacy Models section in the model picker. After that period, it will be retired on June 3, 2026.

FAQs

Q1. What improvements does GPT-5.3 Instant bring compared to GPT-5.2 Instant?

Answer. GPT-5.3 Instant offers smoother, more natural conversations, reduces unnecessary refusals and disclaimers and lowers hallucination rates for more reliable answers.

Q2. How much has GPT-5.3 Instant reduced hallucinations?

Answer. In evaluations, hallucination rates dropped by 26.8% with web access and 19.7% without, making responses significantly more factual and trustworthy.

Q3. Where and how can users access GPT-5.3 Instant?

Answer. GPT-5.3 Instant is available to all ChatGPT users starting today and to developers via the API under the name gpt-5.3-chat-latest. GPT-5.2 Instant remains accessible until June 3, 2026 under Legacy Models.

Also Read

https://www.mymobileindia.com/openai-might-soon-introduce-100-month-chatgpt-pro-lite-subscription-plan/

https://www.mymobileindia.com/openai-to-open-mumbai-and-bengaluru-offices-partner-with-tata-group-for-ai-data-centre-in-india/

Share
Published by
Team My Mobile

Recent Posts

Upcoming Smartphones Launching This Week (April 20–26, 2026) – Huawei Pura 90 Series, Foldable Pura X Max, Motorola Edge 70 Pro and More

Highlights Huawei Launches On April 20 - Huawei will unveil the Pura 90 series with…

9 hours ago

Redmi K Pad 2 Full Specs Leaked Ahead of April 21 Launch, Dimensity 9500 Performance Chipset Performance Leaked

Highlights Redmi K Pad 2 leak reveals an 8.8″ 3K LCD, 165Hz refresh rate, 1100…

9 hours ago

iPhone 18 Pro Colour Options Leaked, the Standout ‘Dark Cherry’ Colour Grabs Attention

Highlights iPhone 18 Pro Colours leak suggests four options including Dark Cherry, Light Blue, Dark…

10 hours ago

Best 5G Phones Under ₹12,000 in India (April 2026) – Redmi A7 Pro 5G, Ai+ Nova 2 5G  and More

Highlights Best 5G Phones - Redmi A7 Pro 5G, Tecno Spark Go 5G, POCO C85…

10 hours ago

Redmi A7 Series India Launch Set for April 21; 6.9-inch 120Hz Display and 6300mAh Battery Confirmed

Highlights The Redmi A7 Series is set to debut in India on April 21, 2026.…

1 day ago

Realme C81 Storage and Colour Options Leaked; Could Launch as Rebranded Realme Note 80 in India

Highlights The Realme C81 surfaces online in 4GB+64GB and 4GB+128GB variants. It will launch in…

1 day ago

This website uses cookies.