OpenAI has announced the global rollout of GPT-5.1, an upgraded version of its flagship AI engine behind ChatGPT, bringing more natural conversational ability, better reasoning and deeper customisation tools.
Two Distinct Modes: Instant and Thinking
GPT-5.1 comes in two variants, Instant and Thinking, designed for different types of interaction. Instant is faster, more playful and better at following user instructions, while Thinking devotes more “thinking time” to complex queries, yielding clearer, more thorough responses.
According to OpenAI, the “thinking time” for this model varies dynamically: for simple tasks it responds very quickly, but for heavier ones it spends more effort to deliver higher-quality reasoning.
A Warmer, More Personal Chat Experience
One of the headline changes in GPT-5.1 is its more human-like tone. The Instant mode now feels warmer by default, thanks to adjustments that make it more conversational without sacrificing clarity.
Perhaps most notably, OpenAI has added a richer set of personality presets for users to choose from. In addition to the original styles, users can now select from eight tonal presets: Default, Friendly, Professional, Candid, Quirky, Nerdy, Cynical, and Efficient.
Beyond that, OpenAI is experimenting with sliders that allow fine-tuning of warmth, conciseness, how “scannable” replies are, and even how often it uses emojis.
Importantly, when you adjust these settings, they apply right away to both new and ongoing conversations, keeping the tone consistent across your chats.
Smarter Reasoning and Better Instruction Following
GPT-5.1 isn’t just about style. OpenAI says the model is better at following instructions, reducing irrelevant or off-track responses, which should make it more reliable for structured tasks like research, writing or technical workflows.
Benchmarks also show gains: in areas such as mathematics and coding, the adaptive reasoning system significantly improves performance. For example, in internal testing on maths evaluations, the new model demonstrated both speed and accuracy improvements.
Tools and Features for Developers
For developers using the GPT-5.1 API, OpenAI has introduced a “no reasoning” mode, which effectively bypasses reasoning effort to handle very straightforward tasks faster.
Two powerful new tools are also being added:
- apply_patch, which allows GPT-5.1 to generate structured code diffs to modify files reliably, rather than just plain text suggestions.
- A shell tool, enabling the model to suggest shell commands that can run on your system, opening up richer agentic workflows.
Moreover, prompt caching has been extended (up to 24 hours), helping save on token consumption when developers reuse prompts frequently.
Rollout and Availability
GPT-5.1 is rolling out gradually. Paid ChatGPT users (Pro, Plus, Go, Business) are first in line. Free and logged-out users will gain access in the coming days.
Enterprise and Education plans have been given a seven-day early-access window, after which GPT-5.1 will become the default model.
OpenAI says that the older GPT-5 model will remain available in a “legacy models” drop-down menu for paid users for approximately three months.
Mixed Reactions from Users
While many users have welcomed the new version as “warmer” and more engaging, some long-time users have expressed concerns. On user forums, a few report that Thinking mode’s quality has declined, with faster responses but more factual errors or “hallucinations.”
One user praised the renewed conversational tone:
“It actually listens again … The way we talk … The way we bounce ideas off each other … That’s exactly the kind of multi-turn, emotionally adaptive … they tuned 5.1 for.”
Another confirmed that personality and custom-instruction updates now synchronise across existing chat threads, not just new ones:
“Personality and custom-instruction changes now apply to all chats immediately, not only to new conversations.”
Looking Ahead
OpenAI describes GPT-5.1 as a meaningful step – not a full generational leap, but a refinement that brings together better reasoning, improved personality, and richer user control.
Moving forward, the company says it plans to expand the personalization controls and continue tuning model behaviour based on user feedback.
Sources: OpenAI
