Blog

The price increase that did not happen

Claude Sonnet 5's introductory $2/$10 was due to rise 50% on September 1. On August 25 that increase was cancelled and the introductory rate became the standard one. We had a countdown clock running to a non-event. Here is what it would have cost, and what we changed.

By the Unbiased Team · published August 25, 2026 · every figure verified against its primary source on August 25, 2026

The 50% increase to Claude Sonnet 5's rates scheduled for September 1, 2026 was cancelled. Anthropic's pricing documentation now states the $2/$10 introductory rate is the standard rate and the increase “will not occur.” Forecasts built on $3/$15 from September are wrong; the metered task that billed $0.0410 in July still bills $0.0410.

What changed

Sonnet 5 launched at $2 per million input tokens and $10 per million output, described as introductory pricing through August 31, 2026, with a standard rate of $3/$15 taking effect September 1. That was a documented, dated, 50% increase, and it was the only scheduled price rise anywhere on the frontier ladder.

Anthropic's pricing documentation now reads: the introductory pricing “is now the standard price” and the previously scheduled increase “will not occur.” No new rate card, no migration, no deadline. The cheap mid tier simply stayed cheap.

What it would have cost

Because both legs were rising by the same 50%, the arithmetic is unusually clean: any input/output mix would have cost exactly 1.5x. Our metered dashboard-generation task billed $0.0410 on Sonnet 5 in July and would have re-priced to $0.0615. A team spending $10,000 a month on Sonnet would have paid $15,000, an extra $60,000 a year.

Slide your own spend through the calculator below to see the bill you were bracing for against the bill you now have. The gap is the size of the reprieve.

We were running a countdown to it

This is the part worth admitting. We built a price-tracker page with a live countdown clock to September 1, and its code was written to flip on the day to a line reading “Sonnet 5 now bills $3 / $15.” Had we not re-checked the source, our own tool would have published a false price change to every visitor, automatically, on a schedule.

Thirteen pages asserted the increase in one form or another: the mid-tier pricing post's entire premise, three comparison timelines, two calculators, the tracker, and the blog index. All corrected on August 25, with the cancellation dated in place rather than quietly deleted.

The mechanism that saved us was not diligence, it was a standing re-verification habit against primary sources. Any page that states a future price is a claim with an expiry date, and unattended claims do not fail loudly. They just become wrong.

The lesson for anyone budgeting model spend

Scheduled price changes are not facts, they are announcements, and announcements get withdrawn. Four dated moves happened on this ladder in ten months and the fifth was called off eight days before it landed. If your capacity planning encodes a future rate, it needs a review date attached, not a calendar reminder that assumes the plan held.

The narrower read is more useful than the general one: the mid tier is now permanently cheaper than the market spent a month planning around. If you moved traffic off Sonnet 5 in anticipation of September, that decision was made on a premise that no longer exists, and it is worth re-running before the quarter closes.

Questions, answered