GPT‑5.1 drops: warmer, sharper, and finally fun to talk to

OpenAI is pushing out GPT‑5.1 starting today, and the headline isn’t just about benchmark gains—it’s about personality. The company is rolling out two variants: GPT‑5.1 Instant, which serves as the everyday workhorse, and GPT‑5.1 Thinking, the heavy-lifting reasoning engine. Paid users on Pro, Plus, Go, and Business plans get first access, with a gradual rollout designed to keep servers from melting down. Enterprise and Edu customers get a seven-day early-access toggle before it becomes the default for everyone.

The big shift with Instant is that the model is now warmer by default. Early testers report it’s surprisingly playful, a marked change from the sterile tone that often plagued previous versions. Beyond the vibes, instruction following has been tightened up significantly, so it answers the question you actually asked rather than the one it thinks you meant. For the first time, Instant can also use adaptive reasoning—deciding on the fly whether to think before responding to hard math or coding problems. That translates to real gains on evals like AIME 2025 and Codeforces.

GPT‑5.1 Thinking isn’t being left behind. It now adapts its thinking time with more precision, spending cycles on genuinely complex problems while speeding through simple ones. The responses are noticeably cleaner, stripped of jargon and undefined terms that made prior reasoning models feel like they were showing off. Its default tone is also more empathetic. As OpenAI puts it, the goal was to make their most capable model actually approachable, especially for complex work tasks.

Alongside the model refresh, OpenAI is refining how you shape ChatGPT’s voice. The old preset options are being revamped: Friendly (formerly Listener) and Efficient (formerly Robot) stick around with updates, while Professional, Candid, and Quirky join the lineup. The idea is to acknowledge that one person’s perfect assistant is another’s annoying sycophant. GPT‑5 Auto will keep routing queries behind the scenes, so most users won’t need to fuss with model pickers—they’ll just notice answers that feel both smarter and more natural. The API gets these models later this week, with Instant as gpt-5.1-chat-latest and Thinking simply as GPT‑5.1. Legacy GPT‑5 models will hang around in a dropdown for three months, giving everyone time to kick the tires before the old guard rides into the sunset.