For years, new ChatGPT models were sold on raw power: more context, better benchmarks, smarter reasoning. With GPT 5.1, OpenAI is finally putting something softer in the spotlight too – how the assistant actually sounds when you talk to it.

Open ChatGPT today and you can pick from personalities like Default, Professional, Friendly, Candid, Quirky, Efficient, Nerdy or Cynical, then tweak sliders for warmth, length, structure and even emoji habit. Those choices apply across models and existing chats, so you set a voice once and the system mostly sticks to it. Under that layer sits the new GPT 5.1 family that actually does the work.

What GPT 5.1 actually is?

GPT 5.1 is an upgrade inside the GPT 5 line, not a fresh generation. OpenAI has split it into two main variants and lets ChatGPT route between them:

• GPT 5.1 Instant for everyday chat, ideas, summarising and quick help
• GPT 5.1 Thinking for heavier reasoning, explanations and multi-step problems

An Auto setting quietly decides which one to use for each query and how much “deliberation” to spend. Instant now uses light adaptive reasoning, so simple prompts stay fast while tougher ones trigger more internal thinking. Thinking adapts its effort more precisely, and OpenAI’s own notes stress that it should answer with clearer, less jargony explanations than earlier reasoning models.

Why it arrives now: backlash, benchmarks and safety?

GPT 5 launched in August as the big flagship, and on paper it was impressive. In practice, the reception was mixed. Coverage from outlets like The Verge and others describes a model that felt strong on tests but oddly formal, to the point where some users asked for GPT 4o back and OpenAI faced enough pushback that it restored older options and promised personality updates.

The GPT 5.1 launch flipped that script in a way the community immediately noticed. One Reddit commenter summed up the release blog with:

“Weird that they didnt include a single benchmark.”

Another added:

“It’s very weird but from what I’ve seen shared on Twitter, it appears to be mostly an upgrade in instruction following and creative writing.”

That mix of scepticism and curiosity is a useful snapshot. OpenAI is clearly leaning into feel and behaviour rather than leaderboard charts, and early users are reading GPT 5.1 as a tone and usability patch on top of GPT 5’s brain.

There is also a quieter safety story underneath. In a new system-card addendum, OpenAI explains that GPT 5.1 Instant and Thinking are evaluated on how they behave in situations involving psychosis, mania or delusions, and on a metric it calls emotional reliance, which covers unhealthy attachment or dependence on ChatGPT. That sits alongside earlier work around sensitive conversations, teen safety and reducing unhealthy emotional reliance in GPT 5, and against a backdrop of lawsuits and investigations about people treating AI chat as an emotional lifeline.

What changes for you in practice?

If you use ChatGPT regularly, GPT 5.1 will feel more like a course correction than a jump. GPT 5.1 Auto is already the default in the app, with Instant and Thinking available directly in the model picker for paid tiers. Pro, Plus, Go and Business users get broader access first, while free accounts see GPT 5.1 within usage limits before the system falls back to smaller models.

The practical playbook is simple: pick a personality that matches how you like to read, nudge the tone and length sliders once, and use Instant for most quick work while switching to Thinking when you genuinely need deeper analysis. Then keep the old rule in mind. GPT 5.1 is a more capable and more approachable colleague, not an oracle – it can still be confidently wrong, even when the tone finally sounds right.

Now, here’s a meme to exactly sum up my original reaction to this launch: