OpenAI replaces GPT-5 with GPT-5.1 in ChatGPT, a more conversational and even smarter model.
And this is a new update for ChatGPT. Three months after the launch of GPT-5, OpenAI is launching GPT-5.1, an improved version of its flagship LLM. Two models were launched this November 12 in ChatGPT: GPT-5.1 Instant and GPT-5.1 Thinking. Auto mode, always present, allows you to automatically choose between the two models. This new version promises more natural conversations with AI, better follow-up of instructions and improved reasoning management. OpenAI promises that GPT-5.1 is smarter as a result.
GPT-5.1, GPT-5.1 Thinking: a change in behavior
To improve GPT-5, OpenAI most likely used RLHF (Reinforcement Learning from Human Feedback), integrating human feedback to refine the tone of the model, as well as reinforcement learning targeted at following instructions. The result is a model with a warmer, more contextual tone, and with significantly enhanced compliance with instructions. The model also progresses in the execution of strict formats, restrictive rules and complex instructions. OpenAI, however, does not publish any precise benchmark to quantify this progress, which is a shame.
Behaviorally, GPT-5.1 now adopts a more natural, more human and more spontaneous response, with better calibrated empathy. Can we deduce from this that the numerous media criticisms of the irregular behavior of ChatGPT pushed OpenAI to intervene on this point? Nothing officially confirms this.
Finally, on reasoning, a point often criticized, OpenAI seems to have adjusted the internal logic of selecting “thinking time”. GPT-5.1 significantly reduces the quantity of reasoning tokens generated for simple tasks, with a drop of around 57%, and on the contrary it increases the depth of reasoning for complex tasks (up to 71% more tokens).
How to personalize the tone of your ChatGPT?
OpenAI also improves ChatGPT’s customization capabilities. The styles allowing you to adjust the way in which the AI expresses itself have been simplified and harmonized, in order to better match real uses. 8 styles are now offered:
- Default (balanced)
- Professional (precise, formal)
- Warm (natural and engaging)
- Spontaneous (direct, honest)
- Quirky (creative, surprising)
- Effective (very concise, minimalist)
- Geek (curious, enthusiastic)
- Cynical (critical, sarcastic)
OpenAI also indicates that GPT-5 will be significantly more rigorous in respecting these styles. The company also announces that it will soon be possible to more finely adjust several parameters: the level of warmth in the tone, the degree of conciseness and the frequency of use of emojis.
Work on context memory
OpenAI finally plans to work harder to personalize the ChatGPT experience for each user. “Users don’t want a multitude of personas or a fragmented personality. They want ChatGPT to feel like a single assistant, able to adapt its tone appropriately depending on the context of the conversation,” explains Fidji Simo. OpenAI plans to make ChatGPT’s memory more intuitive and controllable, so that the assistant appears more attentive without ever becoming intrusive.
Finally, OpenAI number two specifies that this development should not lead to an AI that is too complacent: ChatGPT will have to adjust to everyone, while retaining the ability to challenge the user and help them progress rather than simply proving them right.
