OpenAI has rolled out two updated versions of its AI models, GPT-5.1 Instant and GPT-5.1 Thinking, now integrated into ChatGPT. The launch emphasizes a more anthropomorphic touch, promoting these models as warmer, more conversational, and better at following instructions.
This development responds to earlier criticisms about previous models being overly cheerful and sycophantic. Additionally, it addresses controversy stemming from changes to GPT-5's default style following several suicide-related lawsuits.
As OpenAI navigates legal and regulatory challenges that could impact its operations, the release context has shifted. Unlike past model launches, OpenAI can't simply release new AI models without facing significant scrutiny. Nevertheless, here's what's new: GPT-5.1 Instant is designed for rapid responses, serving as the faster default model, while GPT-5.1 Thinking aims to tackle more intricate problem-solving tasks.
OpenAI reports that both new models outperform their predecessor, GPT-5, in technical benchmarks, including math and coding tests like AIME 2025 and Codeforces.
The main improvement in GPT-5.1, however, is its customizable presentation. OpenAI introduces eight preset communication stylesâProfessional, Friendly, Candid, Quirky, Efficient, Cynical, Nerdy, and the Default settingâallowing users to choose based on their task. These presets adjust the model's instructions for different stylistic effects, while the core functionalities remain consistent.
Image Credit: OpenAI
An illustration presenting GPT-5.1's eight personality styles in ChatGPT.
Furthermore, the GPT-5.1 Instant model incorporates 'adaptive reasoning', enabling it to decide how much computational effort to apply before generating a response.
The roll-out of these models will be gradual over the coming days, beginning with paid subscribers and eventually to free users. OpenAI also plans to offer these models through its API shortly after, with GPT-5.1 Instant and GPT-5.1 Thinking labelled as gpt-5.1-chat-latest and GPT-5.1, respectively, both featuring adaptive reasoning. Meanwhile, the legacy GPT-5 models will stay accessible under a dropdown menu for paid subscribers for the next three months.