The Cheap Tier Just Got Better: Claude Sonnet 5 Is Now the Default

Anthropic released Claude Sonnet 5 on June 30 and made it the default model for Free and Pro users, claiming near-Opus-4.8 quality at a lower price. The per-task cost of the AI workflows your business already runs just dropped.

The customer-support summary your team runs on every closed ticket, the batch of product descriptions your marketing ops person regenerates each week, the classification pass that routes inbound leads: none of these run on the most expensive AI model available. They run on the mid-tier one, because that is where the cost per task is low enough to justify doing it thousands of times. On June 30, Anthropic released Claude Sonnet 5 and made it the default for Free and Pro users, and the mid-tier just got cheaper and better in the same week.

That combination is the story. Not the benchmark. The benchmark is downstream of a simpler fact: the workflows you already pay for got less expensive to run without you changing a line of anything.

What changed

Sonnet 5 is now the default model served to Free and Pro users on Claude, replacing the previous default. Anthropic claims it approaches the quality of Opus 4.8, its top-tier model, on coding, agentic tasks, and knowledge work, while costing a fraction as much to run.

The pricing is the part a business operator should read twice. Through August 31, 2026, Anthropic is running introductory pricing of $2 per million input tokens and $10 per million output tokens. After that, standard pricing settles at $3 input and $15 output. That standard rate is roughly a third lower than the prior Sonnet generation's pricing, and it lands at roughly half the cost of running an Opus-class model for comparable quality on the tasks Sonnet is good at.

As Simon Willison notes in his rundown, the pattern here is the one that has held across recent releases: the quality floor keeps rising while the price of reaching it keeps falling. MacRumors framed it plainly as near-Opus performance at a lower price. That is the whole pitch, and it is an honest one.

Why it matters for a business leader

If you run any AI workflow at volume, your bill is a function of two numbers: how many tokens each task consumes and the per-token price. A default-model upgrade moves the second number down while nudging the first number's usefulness up. You get more correct outputs per dollar, which means fewer retries, fewer human corrections, and a lower effective cost per finished task.

For teams on the paid Claude plans, the upgrade is automatic. The model your staff reach for when they open Claude is now Sonnet 5, at no change in subscription price. For teams building on the API, the change is a config value and a lower invoice.

The strategic read is about margin, not novelty. When the good-enough tier improves, the case for running expensive models on routine work weakens further. A lot of businesses have been paying top-tier prices for tasks that never needed top-tier quality. This release widens the gap between what those tasks require and what the affordable option now delivers.

The honest caveat

Near-Opus is not Opus. For the hardest reasoning, the longest agentic chains, and the tasks where a subtle error is expensive, the top tier still earns its price. Sonnet 5 narrows the gap; it does not erase it. If your workflow is one where being wrong occasionally costs you a customer or a compliance problem, test before you downgrade.

The introductory pricing also has an expiration date. Budgeting a workflow at $2 and $10 per million tokens is fine for a pilot, but the number you should plan production spend against is $3 and $15, effective September 1. Build your cost model on the standard rate, not the promotional one.

And "the default changed" is not the same as "your results changed for the better in every case." A new default model can shift tone, formatting, or edge-case behavior in ways that matter for a tuned workflow. If you have prompts that were carefully calibrated to the previous model, re-check the outputs before assuming the upgrade is free.

The bigger picture

The interesting releases are not always the ones at the frontier. They are the ones that quietly re-price the middle, because the middle is where most real work actually runs. A better top-end model changes what a few teams can attempt. A cheaper, better default changes the cost of what everyone is already doing. This week, the second thing happened, and most businesses will feel it on an invoice before they ever notice it in a demo.