Anthropic’s introductory pricing for Claude Sonnet 5 — the AI model quietly running inside a growing number of business tools, from customer service chatbots to coding assistants to automated research agents — ends on 1 September. From that date, the rate rises from $2 per million input tokens and $10 per million output tokens to $3 and $15 respectively, a 50% increase. It’s the latest reminder that the “AI got cheaper” headlines of the last year sit alongside a quieter, opposite trend: usage-based AI pricing that can move sharply, with the change often landing on the business rather than the end user who barely notices which model sits behind the tool they use daily.

This matters even if you’ve never heard of Claude Sonnet 5 and don’t buy AI models directly. Most SMEs interact with frontier AI models second-hand — through a SaaS product, an agency-built chatbot, a workflow automation tool, or an AI feature bolted onto software you already pay for. Those vendors absorb model costs into their own pricing, and when the underlying model gets more expensive, that cost has to land somewhere: a subscription price rise, a lower usage cap, or a feature quietly moved behind a paywall.

Work out what you’re actually exposed to

Most business owners can name their big software subscriptions but have a much fuzzier picture of which of those tools are billed on usage rather than a flat monthly fee — and usage-based AI tools are exactly the ones that move when a model provider changes its rates. Go through your AI-adjacent subscriptions and ask each vendor directly: is any part of my bill tied to AI model usage, and has that usage cost changed recently or is it expected to? A five-minute email now beats a surprise invoice in October. If you’re not sure which of your tools even qualify, ApplyAI can help UK businesses map out what they’re actually running and where the real costs — and risks — sit, rather than guessing from the invoice line items.

Don’t assume “built for us” means “immune to this”

If you’ve had a bespoke AI tool or agent built for your business — a support bot, an internal research assistant, an automation pipeline — the same exposure applies, just more directly, since you’re often billed on raw API usage rather than a flat SaaS fee. That’s not a reason to avoid custom AI tools; it’s a reason to build them with cost visibility from day one, so a provider’s pricing update shows up as a line on a dashboard rather than a mystery on next month’s statement. BuildApps builds custom AI and app solutions with exactly that kind of cost transparency baked in, so you know what you’re spending and why before the bill arrives, not after.

Check your contract terms, not just the invoice

It’s also worth reading the small print you signed rather than waiting to see what shows up. Some vendors pass model price rises straight through with 30 days’ notice buried in a terms-of-service email; others absorb them for a while and then bundle several increases into one larger annual repricing. Knowing which camp a given supplier falls into changes how you plan — a vendor that passes costs through quickly is more predictable to budget for than one that stores up changes and delivers them all at once at renewal.

The takeaway

You don’t need to switch providers or panic about a 50% rate change on a model you may never interact with directly. You do need to know, this month, which of your tools are quietly billed on AI usage under the hood — because that’s the bill that moves without warning, and the businesses caught out won’t be the ones using AI, they’ll be the ones who never asked how it’s priced.