yesterday (may 5), openai issued an announcement formally upgrading the default model for its chatgpt chatbot to gpt-5.5 instant, replacing the previously used version. the new model emphasizes accuracy and conciseness, particularly by reducing “unnecessary emojis” in responses during user interactions, making conversations cleaner and more efficient.
according to openai, gpt-5.5 instant significantly streamlines response content while maintaining practicality, avoiding lengthy formats and superfluous embellishments. all users will begin receiving the update gradually starting today, while paying subscribers can continue using the original gpt-5.3 instant model as an alternative for the next three months.
in terms of accuracy, the new model has made significant improvements. internal openai assessments show that, for high-risk prompt scenarios such as healthcare, law, and finance, gpt-5.5 instant reduces hallucinatory statements by 52.5% compared with the previous generation. in tests where users actively flagged factually incorrect responses, inaccurate claims also decreased by 37.3%. openai states that the instant series has achieved a comprehensive leap in factual accuracy, with the greatest benefits realized in professional fields where precision is paramount.
this upgrade focuses on making model responses more direct and reliable, addressing issues that previously drew frequent user feedback, such as redundant information and overuse of emojis. industry experts believe this adjustment reflects openai’s ongoing efforts to optimize chatgpt’s practical experience, making it better suited for everyday collaboration in serious contexts. with the change in the default model, free users will directly notice improvements in response quality and speed, while the transition period provided for paying subscribers gives existing users ample time to adapt. openai emphasizes that gpt-5.5 instant represents the optimal balance within the current instant series, combining conciseness, low latency, and high factual accuracy.
