Capable AI Keeps Getting Cheaper. Claude Sonnet 5 Is the Latest Proof

On 30 June 2026, Anthropic released Claude Sonnet 5, which it describes as its most agentic Sonnet model yet, with performance close to that of its flagship Opus 4.8 but at a lower price. Read quickly, that sounds like routine AI news, another model, another set of benchmark numbers. The part worth your attention is the last three words: at a lower price.

Because this keeps happening. Roughly every few weeks, a new model arrives that does more than the one before it and costs less to run. The quality that sat in the expensive top tier a few months ago slides down into the affordable middle, and the floor drops again. Claude Sonnet 5 is simply the clearest recent example: near-flagship capability, offered at the price of the everyday workhorse.

For an Australian small business, the falling price is the headline, not the feature list. It changes the maths on what you can sensibly automate. Work that was too expensive to hand to AI all day, every day, is now cheap enough to run continuously in the background. The question stops being whether you can afford capable AI, and becomes whether your business is set up to use it.

What actually shipped

Anthropic says Sonnet 5 is a substantial improvement over its predecessor on the things that matter for doing real work: reasoning, using tools, coding, and general knowledge work. Early testers reported that it finishes complicated tasks where older models would stop short, and that it checks its own output without being asked. To launch it, Anthropic set an introductory price that runs through to the end of August before standard pricing applies, which is cheaper again than its top-tier model for similar work.

It is not happening in isolation. The same week, there were reports across the industry of the cost of running these models falling sharply as the underlying efficiency improves. The developer and writer Simon Willison, who has tracked the steady collapse in the price of capable AI for years on his blog, has long made the point that today's premium model is next year's bargain. Anthropic's own announcement leans into the same idea, inviting people to dial the effort up or down to find the right balance of cost and performance for each job. The direction of travel is unmistakable, and it points down.

Why a price cut matters more than a new feature

A new feature is exciting for a week. A falling price changes what is possible for good. When capable AI is expensive, you ration it: you use it for the occasional clever task and leave the rest of the business running the way it always has. When the same capability gets cheap, you can let it run across everything, the quoting, the follow-ups, the data tidying, the first draft of every reply, all day, without watching a meter.

That is how the gap between a big company's AI and yours closes. It is not closing because you suddenly get a smarter model than they do. It is closing because the smart model now costs roughly the same for everyone, and the advantage shifts to whoever puts it to work most thoroughly. The binding constraint stops being your budget and becomes your systems: whether you have the plumbing to point capable AI at real work and trust the result.

Where the advantage is

Picture the upside not as one impressive demo but as capable AI quietly earning its keep across the business. Here is what good looks like once the economics are working for you rather than against you:

• Capable AI running across routine work every day, not saved for special occasions because it costs too much to leave on.
• The same kind of automation a large competitor runs, at a price a small business can comfortably absorb.
• A stronger model used where the stakes are high and a cheaper one where they are not, balanced job by job instead of paying top dollar for everything.
• Spend that tracks results and tends to fall as the tools get cheaper, rather than a fixed bill that locks you into one vendor.
• Someone keeping you on the best price-for-performance option as the market moves, so you are never quietly overpaying for last quarter's model.

The price of capable AI keeps falling. The businesses that win are not the ones that pick the perfect model. They are the ones built to ride the curve.NextAura

The practical move is not to chase every launch or agonise over which model is fractionally ahead this month. It is to build your business so it can use whatever is cheapest and most capable today, and swap in the next one without rebuilding everything. We made a related point recently about how AI is becoming the cheapest expert a small business can hire. A falling price is what turns that from a nice idea into something you actually run.

Knowing which model to use for which job, wiring it into the work your business already does, and keeping the whole thing on the best price-for-performance option as the market shifts is exactly the kind of thing we build and run as part of the AI automation we set up at NextAura. The tools keep getting cheaper and better on their own. The advantage goes to the business that is built to take it. If you would rather have capable AI quietly working across your business than spend your evenings reading release notes, get in touch and we will carry it while you get back to running the place.