共有:

The "Silent" Price Hike

Sticker flat.
The bill still climbs.

Nominal token prices haven't moved. Yet the invoice has been creeping upward, month by month. A critical report on Sonnet 5 puts the mismatch into numbers — and it's no longer enough to defend a budget by staring at the price table.

AI Navigate Editorial2026.07.026 min read

STICKER FLAT · TOKENS CREEP UP $/mo Sticker (unchanged) Token usage Apr May Jun Jul
01

The Claim

The price hike
hides on the usage side

The report is straightforward: while the nominal per-token price of Sonnet 5 is unchanged, the token consumption per task has been climbing month over month. Same prompts, but longer answers and more internal reasoning steps — so the invoice quietly grows.

Watch only the price sheet and the trend stays invisible. A real gap opens between nominal and effective pricing, so "we didn't raise prices" and "our bill went up" both end up true at the same time.


02

The Mechanism

Why the tokens
keep creeping up

Several plausible mechanisms line up on the technical side.

SAME PROMPT MODEL INTERNAL More reasoning steps More verbose outputs System prompt bloat BILL ↑
FIG. The same prompt spends more tokens internally, which lands on the invoice.

A critical report says Anthropic keeps nominal Sonnet 5 token prices flat while real costs quietly rise. Multiple factors compound: more reasoning steps, more verbose replies, bigger system prompts, tweaked retry behavior — every one of them lifts token counts.

Anthropic hasn't offered a clear rebuttal so far. From a vendor's perspective, "more tokens" is a byproduct of improving quality; from the paying side, "same operation costs more" is what the ledger says. That's where the accountability tension begins.

03

What Matters

The delta
the price table can't catch

Flat
Nominal token price
Rising
Tokens per task
Both true
"No hike" & bigger bill

Any monthly budget built on token unit price alone misses this shift entirely. Track the average token count for a fixed representative task over time — that single operational habit lets you detect an effective price rise from outside the vendor.

04

Who Feels It

The people running
the forecasts

Finance & PM cost owners

Building next month's budget from the price table stops working. Move to actual-consumption-based forecasting.

Direct API developers

Chart the average token count for a fixed prompt month-over-month — the SDK's token counters make this trivial to instrument.

Light users

At a few dollars a month, the felt impact rounds to noise. Everyday usage stays largely undisturbed.


05

The Frontier

Transparency beyond
the price sheet

This isn't really a narrow argument about pricing. It's about disclosures that let users track effective cost movement — average tokens per fixed task, consumption trends per model generation, breakdowns for internal tool calls. Without that data, "prices unchanged" is functionally the same as promising nothing.

The point applies well beyond Anthropic. The competitive axis is quietly widening from "model quality" to "operational transparency." This report will be worth referring back to as one of the inflection points in that shift.

Source: anthropic.com · AI Navigate — Daily Update · 2026.07.02