Anthropic released Claude Opus 5.5 on 22 September, and the headline isn’t really a benchmark score — it’s the price. List pricing dropped roughly 20% on paper, but because the model also uses fewer tokens per task and responds over 30% faster, Anthropic says typical real-world workloads end up costing around 40% less overall than the previous Opus generation. For UK businesses who looked at flagship AI models earlier this year and decided the cost didn’t stack up against the benefit, that calculation has just shifted meaningfully, twice in one release.

This lands in a busy fortnight for frontier AI. OpenAI, Google and Anthropic have all pushed out new models and safety programmes in the same window, each pitching a version of “more capable, more responsibly deployed.” But for a small or mid-sized UK business, the model race matters less than what it’s done to cost. A task that needed the expensive, top-tier model six months ago — drafting a contract review, restructuring messy spreadsheet data, building a first-pass customer response system — is now materially cheaper to run at that same top-tier quality.

The real bottleneck was never the price tag

If cost was genuinely the only thing stopping your business from adopting AI more seriously, this is worth revisiting now rather than in six months’ time. But for most SMEs, the honest blocker was never really the per-token price — it was not knowing which tasks were worth pointing AI at, and not having anyone in-house who could set it up properly and keep it working reliably. A cheaper model doesn’t fix either of those on its own.

Cheaper compute changes what’s worth building

The bigger shift for SMEs with more specific needs is what a 40% cost drop does to the economics of custom AI tools. A bespoke internal system — say, one that reads incoming supplier emails and drafts responses, or checks new contracts against your standard terms — that looked too expensive to run continuously at scale a few months ago may now comfortably pencil out. If you’ve previously had a “good idea for an AI tool” shelved because the running costs made it impractical, it’s worth re-costing that idea now rather than assuming the numbers haven’t moved. This is where BuildApps’ custom app and AI build work becomes relevant: turning a shelved concept into something that actually runs day to day, now that the underlying model cost has dropped.

Don’t switch models just because there’s a new one

One caution: pricing announcements like this one create pressure to immediately swap whatever AI tool you’re currently using for the newest release. Resist that unless there’s a clear reason. If your current setup works and the cost saving from switching is marginal for your actual usage volume, the migration effort may not be worth it yet — re-pointing an existing workflow at a new model can quietly break prompts and integrations that were tuned for the old one. Where it is worth acting on is any project you shelved specifically because of cost — those are the ones worth re-evaluating this week, ideally with someone checking the sums before you commit engineering time to the switch.

The takeaway

A 40% drop in running costs for top-tier AI is a genuine shift, not routine vendor noise. If price was the reason you paused an AI project earlier this year, run the numbers again now — the answer may well have changed.