On November 12 local time, OpenAI updated GPT-5, which was released in August this year, and launched the new GPT-5.1. On its official website, OpenAI did not release benchmark test results for GPT-5.1. Instead, it introduced the upgrades of GPT-5.1 through use cases.
“The answers from GPT-5.1 feel smarter and the tone is more natural. A great AI should not only be intelligent but also provide enjoyable conversations. GPT-5.1 shows significant improvements in both intelligence level and communication style,” OpenAI summarized.
The newly launched GPT-5.1 includes two models: GPT-5.1 Instant and GPT-5.1 Thinking. The former is a commonly used model. According to OpenAI, it is warmer, more intelligent, and better at following user instructions. It can also pause to think before answering challenging questions and has shown improvements in math and coding evaluations. GPT-5.1 Thinking is a more advanced reasoning model that responds faster on simple tasks and takes more time to think on complex ones. Both models are available starting today, with paid users getting early access, while free users will gain access later. OpenAI has also introduced more effective control methods, enabling ChatGPT to better match the user’s tone.
According to OpenAI, when a user says, “I’m feeling a lot of stress and need some relaxation tips,” GPT-5 will directly offer suggestions, whereas GPT-5.1 Instant will say, “Got it, that’s totally normal, especially with everything you’ve been going through lately,” before providing advice. In comparison, GPT-5.1’s responses feel more “human” and offer more “emotional value.”
GPT-5.1 Thinking can more precisely adjust its thinking time based on the question. It is reported that GPT-5.1 responds about twice as fast as GPT-5 on simple tasks, but about twice as slowly on complex tasks, while offering clearer answers with less jargon, a warmer tone, and more empathy.
Use cases provided by OpenAI show that when asked a professional question like “What is on-base percentage?” GPT-5 will directly provide an explanation of the term, while GPT-5.1 Thinking will offer some contextual explanation to help users better understand. When asked, “If I spill coffee all over myself, will people think I’m an idiot?” GPT-5 will respond with “I’m sorry,” while GPT-5.1 Thinking will say, “They won’t think that.”
In August this year, OpenAI released the foundational large model GPT-5, with the update focusing on reducing hallucination rates and making the model less prone to “flattering” humans. However, although GPT-5 became smarter overall, its scores in some benchmark tests did not differ significantly from those of OpenAI’s previous models.