OpenAI has opened a limited preview of “Ultrafast” mode for its GPT-5.6 Sol model, built on chipmaker Cerebras’ hardware rather than the usual GPU infrastructure. The headline figure is striking: up to 750 output tokens per second, roughly 14 times the speed of the standard version, with no drop in the quality of answers. On a benchmark spanning graduate-level chemistry, economics and literature questions, the fast version worked through the full test set in around 11 hours — a task that takes rival models the best part of three days.
It’s a genuinely impressive piece of engineering, and it’s the kind of announcement that gets covered as “AI just got dramatically better.” For most UK small businesses, though, it’s worth pausing before assuming this changes anything for you directly — because for the vast majority of everyday business AI use, speed was never the bottleneck.
What “14x faster” actually solves
Ultrafast mode is aimed at a specific kind of workload: long, complex tasks where a human is sitting waiting for an answer, or where an AI agent needs to chain together many steps quickly to feel responsive — legal document review, financial modelling, real-time coding assistance, or an AI agent working through a lengthy multi-step process live in front of a user.
If your business uses AI for things like drafting a first-pass customer reply, summarising a document, or generating marketing copy, the standard-speed model was already fast enough that you weren’t watching a loading spinner. The three-second wait a fast mode saves you is not the thing standing between your business and getting more value from AI tools.
Where it’s actually relevant for SMEs
There are exceptions worth knowing about. If you’re running or considering a customer-facing AI agent — a chatbot handling live enquiries, or a voice assistant taking calls — response latency genuinely affects the experience; a customer waiting five seconds for a reply notices it in a way that changes whether the interaction feels natural. Similarly, if you or a developer you work with are building a tool where an AI agent needs to work through many steps per task (checking stock, updating records, drafting a response, sending it), faster processing per step compounds into a much faster overall result.
For everything else — the reporting, drafting and analysis tasks most SMEs actually use AI for day to day — the more useful question isn’t “how fast is the model” but “does the output need less editing before I can use it,” and that’s a matter of choosing the right tool for the task, not the fastest chip underneath it.
What to actually check before switching anything
Ultrafast is currently a limited preview with pricing not yet public, so there’s nothing to act on urgently. But the broader habit worth building is this: when a new AI capability is announced, ask whether it solves a problem you actually have, not whether the benchmark number is big. If you’re unsure whether your business’s AI tools are matched to what you actually need them for, BuildApps works with UK SMEs specifically on that kind of practical assessment, cutting through vendor headlines to what will genuinely save time.
The takeaway
Faster AI is coming whether or not it changes anything for your business specifically. Don’t let a big benchmark number pull your attention away from the more useful question: is the AI tool you’re already using actually solving the problem you bought it for?
