On 7 October, OpenAI pushed GPT-6 into the Chat tab of ChatGPT for every tier, and with it a feature it calls Intelligent UI. Ask a question, and the answer can now arrive as a working calculator, a form, a chart, or a side-by-side comparison, built inside the conversation rather than described in a paragraph. OpenAI's own examples include a retirement savings calculator and a bill splitter. The model is also trained to start answering while it is still reasoning, which OpenAI says gets GPT-6 Instant talking 44% sooner on average than GPT-5.6 Instant on questions that need a web search.
The scale is the main talking point. OpenAI says more than 1.2 billion people use ChatGPT each week, and until this week GPT-6 was for paid customers. We covered the cyber rating on GPT-6 Astra on 8 September, and last week we looked at OpenAI's $100 dots and who carries the can when an always-on agent gets something wrong. This release is a different kind of story: the most widely used AI tool on the planet will now answer a quantity surveyor's question about a cost comparison with something that looks like a spreadsheet.
If your job involves quoting prices to clients, this is where things can go wrong. A calculator with sliders looks polished and tested, even when it hasn’t been.
What OpenAI actually shipped
According to the launch post, GPT-6 has been trained to compose responses from text, visuals, and interactive elements, choosing the mix based on the question. OpenAI built a library of native, streamable components and a compiler that renders the interface progressively as the model writes it. The model decides when a chart beats a sentence and when plain text is enough. OpenAI concedes there is still work to do on the model's design judgement.
The rollout order holds significance for firms with managed accounts. Plus, Pro, Business, and Enterprise tiers got GPT-6 Sol on 7 October. Free and Go tiers got the lighter GPT-6 Luna from 8 October. Enterprise access depends on workplace admin settings, and OpenAI says the models behind ChatGPT Work and Codex are not changing in this release. So if your firm runs ChatGPT Work, your staff may not see this yet. If they also have a personal Free account on their phone, they will.
The 44% faster figure comes from the company’s own tests of web searches, comparing Instant with Instant. It’s an average, not a guarantee. Treat it as an indication, not a proven result.
The math behind the slider
The launch post makes no claim about the numerical accuracy of the calculators the model generates. The system card that accompanies the release does report hallucination evaluations, and OpenAI says GPT-6 Sol and Luna show significant improvements in factuality over their GPT-5.6 counterparts across nearly all metrics, with claim-level errors on factuality-heavy prompts the one exception.
OpenAI also states plainly that those results are from deliberately hard prompt sets and do not reflect production prevalence. One of those sets is described as difficult medical, legal and financial prompts. There is no separate test that I can find of whether a generated cash-flow tool computes the right answer when you move the slider.
The Neuron's Eric Gerard Ruiz and Grant Harvey made the same observation the morning after launch.
OpenAI's announcement makes no accuracy claim for the numbers, so treat a retirement calculator like an intern's spreadsheet and check the formula.
The intern comparison is the useful one. You would not send an intern's first spreadsheet to a client without opening the cells.
They went further on why the format itself is the hazard. "A slider looks like software, and software looks trustworthy, so a wrong number in a polished calculator may land harder than the same mistake in a paragraph."
Anyone who has watched a client accept a tidy Excel output without asking what sits behind it will recognise the problem. The interface does the persuading before the number has earned it.
Gizmodo's AJ Dellinger flagged a second cost. "Intelligent UI could easily burn through people's tokens faster, and visual elements introduce a whole new layer of opportunity for hallucinations." He also noted that Google shipped a similar generative interface with Gemini 3 last year. Google Research's own write-up of that work, from November 2025, admits to occasional inaccuracies in the outputs. Different mechanism, same caveat.
Who signs the number
UK surveyors already have an answer to the liability question, and it has been binding for seven months. The RICS professional standard on the responsible use of AI in surveying practice came into effect on 9 March 2026 for all members and regulated firms worldwide. RICS summarises the requirements under four headings: governance and risk management, professional judgement and oversight, transparency and client communication, and responsible development of AI. The institution's own gloss is blunt. The surveyor remains accountable for every piece of professional advice, regardless of the tools used to produce it.
A note from CMS, written by partner Robert Jones and knowledge counsel Jen White the day after the standard took effect, explains what it means for AI-generated calculators:
Reliability assessment: Where AI output materially affects service delivery, the member must document their assessment in writing, including assumptions, key concerns, mitigations and whether the output can reasonably be used.
Professional accountability: The assessment must be prepared by, or under the supervision of, an appropriately qualified and named surveyor who accepts responsibility for the output’s use.
Terms of engagement: These must disclose when AI is involved and clarify the professional indemnity cover applicable to its use.
Legal implications: CMS notes that a judge or adjudicator is likely to take the standard into account in any proceedings.
Paul Beeston FRICS, Partner and Head of Industry and Service Insight at Rider Levett Bucknall, put the human failure mode in one phrase when RICS gathered practitioners to mark the standard's publication.
I do worry about automation bias – the view that it must be true because the computer says it's true. The standard helps here as it talks about professional scepticism.
Automation bias was a problem when the computer returned a paragraph. A tool that returns a chart with your client's figures already plotted on it makes the bias harder to resist, and the standard gives it a name you will have to answer to.
Where does this land on a live project
Let’s take a hypothetical scenario. A project manager asks ChatGPT to compare the cash flow of two procurement routes for a £12m project. It produces an interactive comparison with a slider to adjust for inflation. The result looks better than the spreadsheet sent by the QS last Tuesday, so the manager takes a screenshot and emails it to the client. Later, nobody can explain which discount rate the model used because the screenshot does not show it.
Some practical consequences follow.
A generated calculator that influences advice is an AI output with a material impact, so somebody named has to own the written reliability decision for it.
Ask for the formula, not just the output, and keep it with the record, because the interface will not survive as evidence in the way a spreadsheet does.
Check which tier your firm's accounts sit on, since Enterprise admins can control this and Work users will not see it yet, while personal accounts already have it.
Rerun the numbers independently before any figure derived from a generated tool reaches a client, and say so in terms of engagement if AI touched the work.
Takeaway
Intelligent UI is a real step in how people will interact with AI. For our sector the shift is narrower and more immediate. A generated calculator is an unaudited model dressed as a finished product, now in the hands of everyone on your project who owns a phone. The RICS standard does not stop anyone using it. It says that the person who relies on the number must be able to show, in writing, why they trusted it. That was always true in principle. From this week it is also the easiest mistake on the project to make.
If you want to short-read a week on what the latest AI releases mean for the people who sign the numbers, subscribe to the Project Flux newsletter at projectflux.ai.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
