GPT-5.1 is the OpenAI model family released on November 12, 2025, in an Instant and a Thinking version. Its idea was effort matched to the question: quick answers to easy questions, more thinking for hard ones. Eleven months later, here is what GPT-5.1 really changed for AI agents and where it stands now, from OpenAI's own pages opened on October 25, 2026.
Ultimo Bots is the AI agent platform where you describe in plain words what your agent should do and Mo, the builder in the AI Studio, builds it: skills that answer from your content and work in your own systems over public APIs and MCP servers, on your website and in WhatsApp, Messenger, Instagram and Slack. You set how thorough it is; which model does the work is our job.
What is GPT-5.1?
GPT-5.1 is a family of OpenAI language models, released on November 12, 2025 in an Instant and a Thinking version, that adapt how long they think to each question. OpenAI's system card addendum, published that day, calls GPT-5.1 Instant more conversational than OpenAI's earlier chat model, with an adaptive reasoning capability that lets it decide when to think before responding. GPT-5.1 Thinking, it says, adapts its thinking time more precisely to each question.
The GPT-5.1 release date is November 12, 2025; the API snapshot carries the next day's date, gpt-5.1-2025-11-13. The GPT-5.1 model page lists a context window of 400,000 tokens, up to 128,000 output tokens, a knowledge cutoff of September 30, 2024, and text and images in, text out, at $1.25 per million input tokens and $10 per million output tokens.
Is GPT-5.1 being removed?
From OpenAI's API, yes, on a schedule. As of October 25, 2026, the GPT-5.1 deprecation is official: the model page shows the status Deprecated, and OpenAI's deprecations page lists GPT-5.1 under October 1, 2026, with removal from the API on April 1, 2027 and gpt-6-sol as the replacement.
Some GPT-5.1 variants went earlier. On April 22, 2026, OpenAI announced the deprecation of gpt-5.1-chat-latest, the API snapshot of the GPT-5.1 model used in ChatGPT, together with the GPT-5.1 Codex models; access was shut down on July 23, 2026. For new work, OpenAI's models overview now says to start with GPT-6 Astra.
GPT-5.2 came after it, and there the launch talk was about self-correction. We sorted the facts from the claims in GPT-5.2 and self-correcting agents.
What changed from GPT-5 to GPT-5.1?
Less than the launch coverage suggested, and nearly all of it about effort and instructions. OpenAI's GPT-5.1 prompting guide names these changes:
- Quicker on easy questions. GPT-5.1 is better calibrated to how hard a prompt is and uses far fewer tokens on easy ones.
- A
nonereasoning setting. Unlike GPT-5'sminimal, it never uses reasoning tokens, for low-latency replies. The model page lists it as the API default. - More steerable. Personality, tone and output format follow instructions more closely.
- Better tool calls. OpenAI calls custom function calling substantially improved, and added coding tools such as
apply_patchand a shell tool.
These are OpenAI's descriptions, not our measurements. The first version of this post turned them into tool-call counts and escalation rates; no source supports those numbers, so this one prints none.
GPT-5.1 Instant and Thinking: effort matched to the question
The idea behind GPT-5.1 Instant and GPT-5.1 Thinking is easy to picture on a business website. "Are you open on Sunday?" needs no deliberation. "Which plan fits a three-person clinic that also wants WhatsApp?" does. A model that thinks the same amount for both is either slow on the first or shallow on the second.
GPT-5.1 makes that call inside the model, question by question. In an AI agent the same trade-off has a price: on Ultimo Bots, a more thorough answer level uses more AI credits per answer. So it is a choice you make on purpose.
What GPT-5.1 changed for AI agents, and what you set yourself
For an AI agent that answers customers, the lasting lesson of GPT-5.1 is that effort and tone should be set on purpose, not left to chance. Here is where each GPT-5.1 change sits in an Ultimo Bots agent, whichever model answers:
Under Intelligence in the AI Studio you choose one level for the whole agent: Ultimo Core at 1 AI credit per answer, Ultimo Pro at 4, Ultimo Max at 8. Core is Fast and economical, for everyday questions, Pro is More thorough, for complex questions and advice. New agents start on Core; Pro and Max need a paid plan.
Tone is yours too. In the AI Studio, the Your agent panel has a Behaviors tab, where Add Behaviors takes a rule of up to 250 characters, such as Answer in two short sentences unless the visitor asks for detail. It is stored word for word and outranks the AI's defaults on every channel.
At 3 AM the level and the rules still apply, and nobody has to be awake. The Talk to a human button on your website shows only while a teammate is online. With nobody there, the agent keeps answering from your knowledge base, takes the visitor's name and email and confirms in its own words that the team has the request. You find the lead with its conversation in the morning.
Is your Ultimo Bots agent ChatGPT?
No. Ultimo Bots is the AI agent platform where you tell Mo, the builder in the AI Studio, in plain words what your agent should do, and Mo builds the skills, including calls to your own systems over public APIs. It is not ChatGPT under another name, and you do not pick a GPT model for it. You choose a level; which model answers on it is set on our side and can change without you rebuilding anything. We do not name the models behind Ultimo Core, Pro and Max. Agents already live keep their content, skills and behaviors when a model changes.
What shapes your agent's answers is in your hands: the content it answers from, the skills it may use and the rules you give it. A price page from last year gives last year's price on any model. Keep the sources behind the agent on your website current, and leave the model question to us.
Plans are priced per agent in AI credits, as of October 29, 2026: Smart $29 a month with 500 AI credits ($19 billed yearly), Boost $59 with 2,000 ($39 yearly) and Ultimo $249 with 10,000 ($199 yearly). On Core, 500 AI credits cover up to 500 answers, and teammates are free. Details on the pricing page and in what an AI agent costs.
When was GPT-5.1 released?
On November 12, 2025, the day OpenAI published the system card addendum for GPT-5.1 Instant and GPT-5.1 Thinking. The API snapshot, gpt-5.1-2025-11-13, carries the following day's date. Instant decides when to think before it answers; Thinking adapts its thinking time to each question.
Is ChatGPT 5.1 being removed?
In OpenAI's API, yes. As of October 25, 2026, OpenAI lists GPT-5.1 as deprecated, with removal from the API on April 1, 2027, and gpt-5.1-chat-latest, the API snapshot of the GPT-5.1 model used in ChatGPT, was shut down on July 23, 2026. For an Ultimo Bots agent nothing changes: you pick a level, and the model behind it is set on our side.
What is adaptive reasoning in GPT-5.1?
OpenAI's term for letting the model decide how much to think. GPT-5.1 Instant decides when to think before responding, GPT-5.1 Thinking adapts its thinking time to each question, and the API adds a none setting that skips reasoning for the fastest replies. Easy questions get quick answers, hard ones more care.
Does GPT-5.1 have voice capabilities?
Not in the API. As of October 25, 2026, OpenAI's GPT-5.1 model page lists text and image input and text output only. With Ultimo Bots, from the Boost plan up, switch on Voice input in the Design tab and visitors can speak their question in any language into the microphone in your website widget or chat link. It appears as text in the input field, they send it, and the agent answers in writing.
How much does an AI agent for a website cost?
Built on GPT-5.1 yourself, OpenAI's API lists $1.25 per million input tokens and $10 per million output tokens, plus hosting and your time, and the model leaves the API on April 1, 2027. With Ultimo Bots you pay per agent in AI credits: Smart $29 a month with 500 AI credits ($19 billed yearly), Boost $59 with 2,000 ($39 yearly), Ultimo $249 with 10,000 ($199 yearly). Every paid plan starts with a free 7-day trial.
Do I need to change my agent when OpenAI retires a model?
No. Your agent answers on the level you picked, Ultimo Core, Pro or Max, and the model behind it is set on our side, so your content, skills and behaviors stay as they are. Worth doing now and then anyway: ask your three most common customer questions in the AI Studio test chat. On Core they use three AI credits.




