OpenAI’s GPT 5.1 Just Landed, and It Talks Like a Friend Who Actually Listens

GPT-5.1 was released by OpenAI today, and the first thing you notice is the tone. Not the voice on your phone, which will be accessible soon, but how it speaks. Say goodbye to those inflexible, textbook responses that made conversing with it feel like talking to a friendly but awkward acquaintance. In its place, you get someone who sounds more like an intelligent buddy, remembering what you said five minutes ago and genuinely asking about your day.

The update is built on two versions: GPT-5.1 Instant, which handles minor daily inquiries, has suddenly taken on a more friendly tone. When you ask for stress-relieving tactics after a bad morning, it addresses you by name, gives you some breathing exercises, and asks how stressed you are. The original approach just provided bullet points and then stopped. This new version stays for a while before checking in and offering to build up a five-minute routine right away.

GPT-5.1 Thinking is for those tough problems. Feed it baseball statistics or a coding puzzle, and it slows down just enough. Simple inquiries still receive a prompt response. The harder ones get the full treatment, with simple phrases replacing the flowery language that used to dominate the screen. Testing suggests that it is now four times faster on easy questions and twice as deep on harder problems than last summer’s GPT-5.

OpenAI seems to have gotten the message. When GPT-5 came out in August, plenty of users just shrugged and thought “yes, its smarter but the personality is still pretty flat” and then took away the older versions all too quick. This time, they’re going slower. First up are paid accounts getting access now, then free users in the coming days, and the old GPT-5 will stick around on a legacy menu for a full 3 months.

Now you can pick from 8 pre-made personality types – Default, Professional, Friendly, Candid, Quirky, Efficient, Nerdy, or Cynical – and each one changes how the answers sound without touching the actual thinking behind the chat. Want direct to the point answers? Efficient does the trick. Need someone to give you some attitude back? Cynical will have a go. You can also tweak the warmth, answer length, emoji use and how often it breaks things down with bold headings. Mess with one setting and all the open chats – including the one you started yesterday – get an instant update.

Instruction following got a major update as well. When you ask for summer travel destinations, it lists five places rather than delving into detail about culture and cuisine. When you push it on a math problem, it pauses, thinks aloud, and then gives you the answer with each step shown. The difference is particularly obvious in protracted talks, as custom rules specified at the beginning continue to apply 20 messages later.

To keep servers stable, access is rolled out in waves, with Pro, Plus, Business and school users getting it first. Enterprise admins get a 7 day test period before it’s locked in. By the end of the week, developers will be able to access both models through the API at the same price as before. GPT-5 will be removed from the main menu in 3 months and 5.1 will take its place. This gives studios, writers and businesses time to test workflows and update prompts. OpenAI will give the same long runway for each subsequent update.