en.Wedoany.com Reported - OpenAI officially announced via its blog on May 5, 2026, that effective immediately, the default model for the ChatGPT chatbot will be replaced from GPT-5.3 Instant to the next-generation foundation model, GPT-5.5 Instant. According to the company's statement, the new model maintains the low-latency characteristics of its predecessor while focusing on improving response accuracy and conciseness. Users engaging in discussions within high-risk, high-sensitivity domains such as medicine, law, and finance will find that the model produces significantly fewer factual errors (the "hallucination" phenomenon), with responses becoming more accurate and reliable. All ChatGPT users have been automatically receiving the update since May 5. Paid users can continue to access GPT-5.3 Instant for the next three months as a transition period, after which the older model will be officially retired.
The core highlight of this update lies in the significant improvement in accuracy. Internal evaluation data released by OpenAI shows that when processing prompts covering high-risk sensitive domains such as healthcare, law, and finance, GPT-5.5 Instant produced 52.5% fewer "hallucinated statements" compared to GPT-5.3 Instant. Concurrently, in tests involving conversations where users flagged factual errors, the number of inaccurate statements output by the new model decreased by 37.3%. OpenAI summarized in its announcement that the Instant model series has achieved comprehensive progress in factual accuracy, with the most significant gains realized in domains where accuracy is most critical.
GPT-5.5 Instant also outperformed its predecessor across multiple high-credibility benchmarks. According to the company's announcement and reports from tech media outlet SiliconANGLE, on the HealthBench medical Q&A dataset, GPT-5.5 Instant's score improved from GPT-5.3's 49.6 points (out of 100) to 51.4 points; on the HealthBench Professional version oriented towards clinical application scenarios, the score jumped from the predecessor's 32.9 points to 38.4 points. OpenAI emphasized that given health is one of the most common topics users discuss with ChatGPT, reliability improvements in this domain are crucial. Furthermore, the new model achieved a score of 81.2 points on the AIME 2025 mathematics test (compared to the predecessor's 65.4 points) and improved from 69.2 points to 76 points on the MMMU-Pro multimodal reasoning benchmark, demonstrating synchronous enhancement in general task capabilities.
Beyond accuracy, GPT-5.5 Instant also achieves two major optimizations in response style and information management. On one hand, OpenAI focused on reducing response verbosity; the new model will tend to provide more compact, to-the-point answers and reduce the use of "unnecessary emojis," making communication more efficient. On the other hand, the new model has received significant enhancements in memory and personalization features. For Plus and Pro subscribers on the Web version, GPT-5.5 Instant can utilize built-in search tools to look back through past chat history, user-uploaded files, and Gmail content, thereby providing more customized and precise personalized answers. The rollout on mobile is also planned and will gradually extend to Free, Go, Business, and Enterprise plan users in the following weeks. OpenAI also disclosed that to enhance data transparency and user control, all ChatGPT models now feature a "Memory Sources" function, allowing users to understand which background information the platform referenced to generate personalized responses, and granting them the right to delete outdated sources or correct erroneous information, although shared chat content remains invisible to recipients.
OpenAI co-founder and President Greg Brockman previously commented on the GPT-5.5 series, stating that compared to GPT-5.4, this series operates faster, features more precise logic, and consumes fewer tokens, making more cutting-edge AI capabilities accessible to both enterprises and general users, which is one of the team's goals. Analysis suggests that the debut of GPT-5.5 Instant represents the efficiency-prioritized version within the series, with its design logic stemming from user feedback expressing annoyance at meaningless polite phrasing. This adjustment direction aligns with the consensus among multiple artificial intelligence companies on enhancing the practicality and trustworthiness of AI assistants.
This upgrade integrates high accuracy, low latency, personalized service, and factual transparency into a single default experience, covering the daily interaction entry point for hundreds of millions of users, and allows developers to call the GPT-5.5 model via the API using the "chat-latest" label. During the three-month transition period, enterprise users can still switch between the old and new models to complete business adaptation.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
