Home Services Work About Blog FAQ Contact
← Back to blog

Anthropic just canceled the Claude Sonnet 5 price hike.

Anthropic said Claude Sonnet 5's $2/$10 API pricing was introductory, good through August 31. We wrote in June that the bill would jump on September 1. It didn't — and the reason why matters more than the number.

On August 10, Anthropic quietly added a note to its pricing docs: the increase to $3 per million input tokens and $15 per million output tokens, originally scheduled for September 1, "will not occur." The $2/$10 rate for Claude Sonnet 5 is now the standard price, not a promo that's about to expire.

We flagged the opposite outcome back in June, when Sonnet 5 launched. The math we ran assumed the discount would end on schedule, the same way every other Anthropic pricing window has ended on schedule. This time it didn't, and the reason is worth digging into — it changes how you should read any "introductory pricing" line from any AI vendor, not just this one.

What Anthropic actually said

Here's the note, word for word: "The $2/$10 per million input/output token pricing for Claude Sonnet 5, announced at launch as introductory pricing through August 31, 2026, is now the standard price. The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur."

No blog post. No press release. One sentence, dropped into a pricing table most people only open when they're already worried about a bill. That's usually how the boring-but-important changes get communicated, which is exactly why most teams running Sonnet 5 in production haven't noticed yet.

Why this almost never happens

Vendors don't usually walk back a scheduled price increase. They extend a promo. They grandfather existing customers quietly and raise the rate for everyone new. Canceling the increase outright and calling the discounted rate "standard" is a different move — it's Anthropic telling the market, whether they meant to or not, that $2/$10 already works at scale.

A canceled price increase is a vendor telling you what the market did to their margin, whether they meant to or not.

There's a specific reason this one matters beyond the headline number. Sonnet 5 uses a newer tokenizer that produces roughly 30% more tokens than Sonnet 4.6 did for the same block of text. At $3/$15, that tokenizer change alone would have pushed Sonnet 5's real-world cost about 30% above Sonnet 4.6's old rate, even though the nominal per-token price would have looked identical. Locking in $2/$10 instead keeps Sonnet 5 close to a wash against 4.6 — arguably cheaper, depending on your workload. Not an accident.

What it does to your bill

Take the lead-qualification workload we used as a baseline in our agent cost breakdown: about 3,700 tokens per conversation, 10,000 conversations a month. At $3/$15, once you account for the extra tokens the new tokenizer generates, that workload lands somewhere north of $650 a month. At the now-permanent $2/$10, it's closer to $430. That's not rounding-error money — it's the difference between a line item you have to defend in a budget review and one nobody asks about.

If you built a Sonnet 5 workflow this summer and budgeted for the September jump, put that buffer back to work. If you shelved a build because the "real" price looked too steep once the promo ended, that reason is gone.

The permanent rate carries through the rest of the pricing table too, not just the headline number. Cache reads on Sonnet 5 sit at $0.20 per million tokens — a tenth of the base input rate — so a workflow that reuses the same system prompt or document context across calls gets cheaper faster than the sticker price suggests. That math worked before the cancellation. It's just easier to plan around now that the base rate isn't scheduled to move out from under it.

What to do with this

Three things, this week, not eventually:

We build and run these agents for clients, so we watch pricing pages the way some people watch stock tickers — mostly out of self-interest, honestly. This is one of the rare weeks it worked out in your favor without you having to do anything. Our original Sonnet 5 breakdown covers the full model comparison if you're still choosing between tiers.

Want a second opinion on what a specific workflow actually costs to run at current pricing? Send us the details and we'll run the numbers before you commit to anything.

— Cole

Sources

Not sure what your AI workflow actually costs at current pricing?

30-minute call. We'll run the real numbers for your use case — no pitch, just math.

Book a Discovery Call →