OpenAI quietly activated GPT-5.1 this week, three months after GPT-5 debuted. “Warmer, smarter, and better,” said the company in comments reported by PCMag.
CNET reported that the update aimed to help ChatGPT communicate in a more natural, conversational way and to better understand instructions.
The company unveiled two variants—GPT-5.1 Instant and GPT-5.1 Thinking—as refinements to the GPT-5 family released in August. Instant replied faster and suited back-and-forth chats, while Thinking spent more time on complex problems yet still responded quickly to simple ones, a method the firm called adaptive reasoning. OpenAI said both versions used less jargon and fewer undefined terms when topics turned technical.
Personalization shaped the launch. Users now choose among eight communication styles: Default, Friendly, Efficient, Professional, Candid, Quirky, Cynical, and Nerdy. The model’s capabilities remained the same across options, but the tone shifted from formal to familiar, playful to direct, or jargon-heavy to plain.
Beyond tone, sliders let users set formality, humor, and emoji frequency, an effort to make replies sound less robotic and more personal. “We want ChatGPT to feel like yours and work with you in the way that suits you best,” said Fidji Simo, OpenAI’s chief of applications, in a company blog post. She added that with more than 800 million people using the service, the firm replaced one-size-fits-all menus with simple, guided control.
GPT-5.1 reached Pro, Plus, Go, and Business subscribers first. Enterprise and Edu customers received a seven-day early-access toggle, off by default, before the model became their standard option. Free and signed-out users would gain access later. GPT-5 remained in the legacy dropdown for three months, part of a sunset period introduced after earlier backlash over model removals.
OpenAI said GPT-5.1 outperformed GPT-5 on math and coding benchmarks such as AIME 2025 and Codeforces, processed requests faster, handled natural language better, and followed instructions more closely. One demonstration showed the model limiting an answer to six words, a constraint earlier versions often ignored.
“A nice upgrade,” said Sam Altman, OpenAI’s chief executive, in remarks reported by PCMag. He pointed to stronger instruction following, adaptive thinking, and improved style, noting that GPT-5 had felt less friendly than GPT-4o and that the new release should seem warmer without annoying most users.
Adaptive reasoning let the system decide how deeply to think on the fly. Simple prompts routed to Instant; thornier requests triggered Thinking. For those who preferred not to choose, GPT-5.1 Auto examined the request and selected a path.
The company addressed safety concerns about unhealthy attachments to chatbots. “We have to be vigilant about the potential for some people to develop attachment to our models,” said Simo, adding that OpenAI worked with mental-health clinicians to define healthy, supportive interactions.
Legal challenges persisted. OpenAI fought a court order to turn over 20 million ChatGPT conversations, arguing that the ruling disregarded privacy protection and broke with common-sense security practices, while Magistrate Judge Ona Wang noted that the firm had not explained why existing safeguards were insufficient.
Paid subscribers now see GPT-5.1 Instant, GPT-5.1 Thinking, and an Auto router in the model picker; free-tier users still see GPT-3.5 until the wider rollout finishes. The company said future models would arrive with similar comparison periods and that sunset timelines would be communicated clearly.
Produced with the assistance of a news-analysis system.