AI Models Keep Getting Cheaper. Here's What That Means for You.

Anthropic released Claude Opus 5 on July 24. On its own, a new model version isn't blog-worthy — there's a new one every few months from somebody. What's worth noticing is the specific trade it makes: according to Anthropic's own announcement, Opus 5 performs close to the company's most capable model on coding and knowledge-work benchmarks, at half the price of reaching that tier before, while its per-token pricing stays the same as the previous Opus release. The model is rolling out across Anthropic's consumer and business plans as well as the major cloud platforms that resell it.

Why one model update is worth a blog post

I'm not writing this to tell you to switch tools. I'm writing it because this is the third or fourth time in the past year that "the good AI model" has gotten meaningfully cheaper or more capable without the price going up. That's a pattern, not a one-off, and it matters more to a small business than any single release does.

If you looked at AI tools a year ago and decided they weren't worth the cost or weren't good enough for real work, that evaluation has an expiration date. It's not that every business needs to redo that evaluation every quarter — most don't have the time, and chasing every release is its own waste of time. But an evaluation that's a year old is evaluating a product that doesn't exist anymore.

What to actually do with this

Don't rebuild around one vendor's model. The pace of change means whatever's "best" now won't hold that position long. Favor tools and workflows that aren't locked to a specific model version.
Re-check cost, not just capability. A task you ruled out for being too expensive to run through AI six months ago may pencil out differently now — the price-per-quality-of-output ratio keeps moving in one direction.
Re-test the specific task, not the model in general. Benchmarks measure broad capability; whether a model handles your invoice descriptions or your customer email backlog well is a five-minute test, not a research project.

None of this means every business needs an AI tool right now. It means "we looked into it and it wasn't there yet" is worth revisiting periodically, on a schedule that fits your business — not because the hype cycle says so, but because the underlying economics keep shifting.

Sources: