Claude Sonnet 5 costs $2 per million input tokens and $10 per million output tokens, with cached input at $0.20 and cache writes at $2.50 for the five-minute window or $4 for the hour, read from the Anthropic pricing page on 16 August 2026.
Those rates were announced at launch as introductory pricing through 31 August 2026, and Anthropic has since confirmed they are now the standard price: the increase to $3 and $15 scheduled for 1 September will not happen. Sonnet 5 is therefore cheaper than the Sonnet 4.6 it replaced.
What Claude Sonnet 5 costs, line by line
Anthropic bills on five lines rather than two. The two cache write columns are what most calculators leave out, and on a model this cheap they are also the lines people ignore hardest.
| Billing line | Sonnet 5, per million tokens |
|---|---|
| Input | $2.00 |
| Cache write, 5 minute window | $2.50 |
| Cache write, 1 hour window | $4.00 |
| Cached input read | $0.20 |
| Output | $10.00 |
Sonnet 5 at $2 in and $10 out per million, read 16 August 2026
The cache lines are fixed multiples of input across the Claude range: 1.25x for a five-minute write, 2x for an hour, and 0.1x to read either. Two cases break the read multiplier: Claude Fable 5.1 and Claude Mythos 5.1, where a read costs 0.025x (a ratio Fable 5.2 and Mythos 5.2 kept when Anthropic listed them on 9 November 2026), and Claude Opus 5.5 and Claude Sonnet 5.5, where it costs 0.05x, per the Anthropic pricing page read on 29 September 2026, with the Sonnet 5.5 cache-read row re-read on 13 October 2026 after Anthropic’s 7 October price cut. On Sonnet 5 that puts a cached read at twenty cents per million, a floor it shares with Opus 5.5. Claude Sonnet 5.5, which the same page lists at the same $2.00 and $10.00 as Sonnet 5, now reads its cache at ten cents per million since Anthropic cut that line on 7 October 2026.
The generation that got cheaper
A new model priced below the one it replaces is rare enough that it is worth putting on a chart.
| Model | Input / cached / output, per million |
|---|---|
| Claude Sonnet 5 | $2.00 / $0.20 / $10.00 |
| Claude Sonnet 4.6 | $3.00 / $0.30 / $15.00 |
| Claude Sonnet 4.5 | $3.00 / $0.30 / $15.00 |
There is a catch that does not appear on the grid, and it matters more than the headline discount. Anthropic notes that Claude 4.7 and later models use a newer tokenizer that produces roughly thirty percent more tokens for the same text. Sonnet 5 is on that tokenizer; Sonnet 4.6 and earlier are not.
So a third off the rate, against roughly thirty percent more tokens billed for identical input. On text-heavy workloads those two move in opposite directions and can very nearly cancel out. The honest way to compare is to run your own corpus through both and count tokens, not to compare the per-million rates and stop there.
A third off the price and thirty percent more tokens for the same text is not a third off the bill. On some corpora it is close to no change at all.
The increase that did not happen
Between the launch of Sonnet 5 and August 2026, the published plan was that $2 and $10 were introductory and would rise to $3 and $15 on 1 September 2026. Anthropic has since cancelled that increase and made the launch rates standard. Here is what the difference would have been on a modest month.
Fifty percent on every bill, for as long as you would have kept running. We mention this less as history than as a caution about how these pages should be read: a rate with a published expiry is not a rate, it is a rate and a date, and both can move. Ours carry the day they were read for exactly that reason.
What a month actually costs
Three shapes of work, with ninety percent of input served from cache, which is realistic for anything carrying a stable system prompt.
| Monthly workload | Sonnet 5 / Haiku 4.5 / Opus 5 |
|---|---|
| Assistant: 2M in, 0.5M out | $5.76 / $2.88 / $14.40 |
| Same volume, no caching at all | $9.00 / $4.50 / $22.50 |
| Agent: 20M in, 4M out | $47.60 / $23.80 / $119.00 |
Sonnet 5 sits at exactly twice Haiku 4.5 and 2.5 times below Opus 5, at every volume, because the whole grid is scaled rather than the individual lines. That makes the tier decision cleaner than it usually is: your work either needs the tier above or it does not, and the answer does not change when you grow. Our LLM cost calculator prices the same volume across every provider we track.
Cheaper ways to run the same model
Two levers, and they stack. The Batch API takes fifty percent off input and output alike, putting Sonnet 5 at $1.00 and $5.00 per million, which is Haiku money for a Sonnet answer, at the cost of running asynchronously. Prompt caching takes ninety percent off any input you resend, which we work through in detail on our Claude API pricing page, cache included.
Neither changes the model you get. Together, on a workload that repeats a large system prompt and tolerates a queue, they are the difference between a bill you watch and one you forget about.
When Sonnet 5 is the right tier
Sonnet 5 is the default for production work: cheap enough to run at volume, capable enough that most tasks do not need the tier above. The two reasons to leave it are specific. Move up to Opus 5 at 2.5 times the price when a wrong answer costs more than a month of inference, or when you would otherwise run several Sonnet passes and a judge to get one usable result. Move down to Haiku 4.5, at half the price, for classification, extraction and routing, where the task is narrow and the volume is large.
Below Haiku the market keeps going, and it goes a long way: we mapped where the floor actually sits across every provider we track.
Frequently asked questions
How much does Claude Sonnet 5 cost per million tokens?
$2.00 for input and $10.00 for output, with cached input at $0.20 and cache writes at $2.50 for five minutes or $4.00 for an hour. Read from the Anthropic pricing page on 16 August 2026.
Is the Sonnet 5 introductory price still going to expire?
No. Anthropic states that the $2 and $10 rates announced as introductory through 31 August 2026 are now the standard price, and that the increase to $3 and $15 scheduled for 1 September 2026 will not occur.
Is Sonnet 5 really cheaper than Sonnet 4.6?
On the rate, yes, by a third on every line. On the bill, not necessarily: Sonnet 5 uses a newer tokenizer that produces around thirty percent more tokens for the same text, so the two effects partly cancel. Measure your own corpus before assuming a saving.
What is the cheapest way to run Sonnet 5?
The Batch API, at $1.00 input and $5.00 output, a flat fifty percent off, stacked with prompt caching at a tenth of fresh input. The cost is that batch requests are asynchronous.
Sonnet 5 or Haiku 4.5?
Haiku 4.5 is exactly half the price on every line. It is the right call for narrow, high-volume tasks such as classification and extraction, and the wrong one for anything where the model has to hold a long thread or make a judgement.
What we verified and what we did not
Verified: all five billing lines for Sonnet 5, 4.6 and 4.5, the note confirming the introductory rate is now standard and the September increase cancelled, the tokenizer note on Claude 4.7 and later, the fifty percent Batch API discount and the cache multipliers, all read from the Anthropic pricing page on 16 August 2026.
Not verified, and therefore not asserted: how much the newer tokenizer actually adds on your own text, which depends entirely on your content and which we have not measured; rates on Bedrock or Google Cloud, which publish their own grids; and any benchmark or quality claim, since this page prices models rather than ranking them.
Prices in this category move, and this one was scheduled to move before the plan changed. Everything here carries the date it was read, and the grid is worth checking at the source before you commit a budget to it.