Google released Gemini 3.7 Flash this week, barely three weeks after its predecessor, and did it with an introductory offer that halves the usual API price until the end of the year. It’s aimed squarely at coding, agentic workflows, and document processing — exactly the kind of tasks SMEs are increasingly asking AI tools to handle. Benchmarks show a real jump in coding performance over the previous version too, not just a cheaper price tag on the same capability.

This is part of a pattern that’s been building all year. OpenAI cut GPT pricing earlier this month, other providers have followed, and the general direction of travel for the models doing the heavy lifting behind everyday business AI tools is down. If you run a business that already uses AI-powered software, or you’ve been putting off adopting it because the running costs felt uncertain, this is worth ten minutes of attention.

Cheaper doesn’t mean irrelevant to your bill

It’s tempting to read “price halved” as “AI just got free” and move on. It didn’t, and the detail matters. Gemini 3.7 Flash’s discount is introductory — list prices are already scheduled to roughly double again from January 2027. Providers are using aggressive pricing to win developer mindshare while the market is still being decided, not committing to permanently cheap AI. Any cost planning that assumes today’s price holds indefinitely is planning on sand.

What’s actually happening is that the unit cost of AI processing is falling roughly in step with rising capability, which is different from AI becoming free. Most SMEs don’t pay these API prices directly anyway — you pay for the SaaS tool built on top, whether that’s a customer service assistant, a document processor, or a coding copilot embedded in your dev tools. Falling underlying model costs eventually show up as better margins for those vendors or lower subscription prices for you, but rarely both, and rarely immediately.

What this actually means for your business

If you’re already paying for AI tools, this is a good moment to ask your vendor a pointed question. Underlying model costs have fallen sharply this year. If your subscription price hasn’t moved, ask why, and whether a renewal is coming up where you can push back.

If you’ve been holding off on adopting AI tools because of cost uncertainty, the calculus has shifted in your favour. Tasks that were borderline cost-effective six months ago — bulk document summarisation, first-draft customer replies, code review assistance — are now meaningfully cheaper to run, and the newer models are also more capable, which means fewer retries and less human cleanup per task.

Don’t chase every new model release. The three-week gap between Gemini 3.6 and 3.7 Flash is a sign of how fast this space is moving, not a signal that you need to re-evaluate your stack every month. Pick a tool that fits your workflow, judge it on results, and let your vendor worry about which underlying model powers it.

If the tooling you need doesn’t exist off the shelf, BuildApps builds custom AI-powered applications designed around your actual workflow rather than a generic one.

The takeaway

AI is getting both cheaper and more capable at the same time, and that trend looks set to continue through the rest of 2026. That’s genuinely good news for SME budgets, but the value only lands if you’re actively checking whether your current tools and vendors are passing the savings on — falling API prices don’t automatically show up in your invoice unless someone asks.