OpenAI officially announced today that its latest model GPT-5.5 Instant has been opened to paying users around the world and will be fully rolled out to all free users on the 26th.This is OpenAI’s most extensive product upgrade since entering 2026.GPT-5.5 Instant will directly replace the original GPT-5.3 Instant and become the default model for the entire ChatGPT platform.Hosts the daily conversations of hundreds of millions of active users.
In terms of performance, official data from Open AI shows that the hallucination rate of the new version of the model in high-risk fields such as medical care, law, and finance has dropped by 52.5% compared with the previous generation GPT-5.3 Instant. In conversations that users actively marked as factual errors, inaccurate statements have been reduced by 37.3%.
In hard benchmark tests, the accuracy of the AIME 2025 mathematics competition jumped from 65.4% to 81.2%, the GPQA doctoral-level science questions increased from 78.5% to 85.6%, and the MMMU-Pro multi-modal reasoning increased from 69.2% to 76.0% - many indicators are close to the level of the flagship model two years ago.
In terms of user experience optimization, GPT-5.5 Instant reduces redundant politeness and excessive segmentation, making the output more direct and clearly structured.With the same amount of information, the number of words is reduced by about 30%, and the number of lines is reduced by nearly 29%.The lengthy lists, excessive paragraphing, and redundant polite words that were widely criticized before have been significantly reduced, and the responses are more direct and the structure is clearer.
The new version of GPT-5.5 Instant simultaneously optimizes the accuracy of image understanding, STEM question answering and Internet search judgment., the completion of high-frequency scenarios such as daily information inquiry, operation guides, technical writing, and translation has been significantly improved.
More importantly, the model has a built-in intelligent routing mechanism that can automatically determine the complexity of user problems. Simple tasks maintain low-latency responses, while complex tasks will silently switch to the Thinking deep reasoning mode in the background. Users can obtain matching capability output without manual switching.
As for the context window, hierarchical configuration is implemented:16K for free users, 32K for Plus and Business users, and 128K for Pro and Enterprise users.This differentiated design not only maintains the basic usability of the free version, but also leaves a clear value anchor for paid upgrades.