OpenAI Group PBC has announced the first major upgrade to its flagship GPT‑5 large language model, following mixed reviews when it launched in August.

The update is intended to make ChatGPT “smarter and more enjoyable” to converse with, the company said today. The release introduces two variants called GPT-5.1 Instant and GPT-5.1 Thinking.

The former, which will replace GPT-5 as the default in ChatGPT, is said to be “warmer, more intelligent and better” at following user’s instructions than GPT-5. OpenAI says the latter is a smarter version of the model that’s meant to handle more complex queries and tasks, and is “easier to understand and faster on simple tasks,” while being “more persistent” when handling complex tasks.

“We heard clearly from users that great AI should not only be smart, but also enjoyable to talk to,” the company said in a blog post announcing the models. “GPT-5.1 improves meaningfully on both intelligence and communication style.”

Enhanced reasoning

In a separate blog post on Substack, OpenAI Chief Executive of Applications Fidji Simo said the major difference between GPT-5 and GPT-5.1 is that the models now converse in a more natural tone. “It often surprises people with its playfulness while remaining clear and useful,” she promised, citing early user feedback.

GPT-5.1 Instant is able to leverage newfound “adaptive reasoning” capabilities to determine when it needs to think more carefully about its responses, such as when someone enters a more complex prompt. Improvements have also been made to its instruction following skills, so it will directly address user’s queries with more consistency than before.

When OpenAI first launched GPT-5, CEO Sam Altman made a lot of noise on social media, hyping up its capabilities, but many users were disappointed once they saw it for themselves. The model was criticized for introducing only incremental improvements, and many were upset that OpenAI decided to make it the default model in ChatGPT while removing earlier models. The backlash was so intense that the company one day later opted to reinstate GPT-4 as an option.

GPT-5 also struggled in terms of performance, with more recent model releases such as Baidu Inc.’s ERNIE-4.5-VL-28B-A3B-Thinking ranking higher in key benchmarks such as instruction-following.

GPT-5.1 Thinking is meant to address that, and OpenAI said it gains the ability to work out how much “reasoning power” it needs to devote to a prompt, based on its complexity. As such, it will take longer to respond to more complicated queries than simple questions. However, OpenAI said early tests demonstrate that the model spends less time on average and uses fewer tokens than GPT-5 on simpler tasks, while also delivering better-quality responses.

The company added that GPT-5.1 Thinking’s responses will also come with “less jargon” and use fewer “undefined terms” than before. The idea is to make the model more approachable and easier to understand when explaining technical concepts.

Greater customization

The other main focus of the new models is on personalization. Users can switch between various different modes that adjust their tone, depending on if they want a friendlier chat experience or a more authoritative voice.

Though ChatGPT did already offer some basic presets for model tone, it has expanded these options. Users can now select from “default,” “friendly,” “efficient,” “cynical,” “nerdy,” “professional,” “candid” and “quirky” modes, the company said. In addition, users will also be able to adjust how often ChatGPT uses emojis.

Simo explained that the company has tried to strike a better balance between GPT-5.1’s IQ and EQ, but realized that everyone has their own preferences regarding tone, hence the expanded range of presets. “The model has the same capabilities whether you select default or one of these options, but the style of its responses will differ — more formal or familiar, more playful or direct, more or less jargon or slang,” she said.

OpenAI will be very interested in user’s initial feedback, because it’s facing pressure from all sides as the artificial intelligence race intensifies. The company suffered a setback recently, with the stinging criticism of GPT-5 persuading Microsoft Corp. to look at alternative models from one of OpenAI’s biggest rivals, Anthropic PBC. That company’s models are now available as an option in Microsoft’s Copilot Researcher, Copilot Studio and GitHub Copilot, as well as its new Office agent.

The models will be made available initially to ChatGPT Pro, Plus, Go and Business subscribers, and will also come to the free version of the chatbot. The company plans to roll out GPT-5.1 gradually to keep performance stable, so not all users will see it right away. Meanwhile, the original GPT-5 will remain an option for ChatGPT users for the next few months.

Images: OpenAI