Guides / claude opus 4.8 pricing

Claude Opus 4.8 pricing, and why the same rate isn't the same cost

Claude Opus 4.8 pricing is $5.00 per million input tokens and $25.00 per million output tokens, matching Opus 4.5 through Opus 5. The rate is identical and the bill isn't, because Claude 4.7 and later use a newer tokenizer. Culpa, a local-first LLM cost, margin, and forecast ledger, meters the tokens each version actually produced rather than the rate it published.

Why this happens

Five Opus versions share one rate in this price book, Opus 4.5 through 4.8 plus Opus 5, and an earlier version of this page concluded from that an Opus version pin saves nothing at all. That was wrong, and the correction is the most useful thing here. Anthropic's own pricing page states that Claude 4.7 and later models use a newer tokenizer which produces approximately 30% more tokens for the same text, with the exact increase depending on content and workload shape, while Claude Sonnet 4.6 and earlier use the previous one. Opus 4.5 and 4.6 sit on the old tokenizer and Opus 4.7, 4.8 and Opus 5 sit on the new one. So the rate card is identical across all five and the bill isn't, because the same prompt becomes a different number of billable tokens depending which side of that line you're on. A pin at 4.6 really can be cheaper than 4.8 on identical text. What it buys is a token count, not a discount, and the only way to know the size on your own traffic is to meter both.

What this usually looks like

  • Production is pinned to an Opus version from several releases ago and nobody remembers deciding that.
  • The pin is justified from the rate card, which is the one place the answer isn't.
  • Different services run different Opus versions and nobody has compared their tokens per request.
  • The prompt attached to the older pin is noticeably longer than the one used elsewhere.

Free, no card, no account

Run the free Cost Leak Scan

It shows your most expensive conversation before you install anything.

Run the free Cost Leak ScanStart 14-day trial

Mistakes that cost the most

MistakeWhy it hurtsDo instead
Reading an identical rate card as an identical bill.Claude 4.7 and later tokenize the same text into roughly 30% more tokens, so the rate matches and the invoice doesn't.Send the same prompt to both versions and compare the reported input tokens, not the rates.
Carrying an old prompt forward across an Opus upgrade.Prompts written for an older version often over-steer, and on 4.7 and later that fatter prompt also tokenizes higher.Re-test whether the extra steering is still needed, then trim what a newer version handles unprompted.
Running the whole agent loop on Opus because the reasoning step needs it.Routing and formatting steps bill at $5 and $25 per million to do work a small model does identically.Split the loop by step type and price each. This is usually the largest single saving on an agent.

Run this check tonight

  1. List which Opus version each service pins, and confirm anyone can explain why.
  2. Send one identical prompt to Opus 4.6 and Opus 4.8 and compare the input token counts they report.
  3. Compare the system prompt on your oldest pin against your newest. Price the difference at $5 per million.
  4. Count what share of your Opus calls are reasoning steps versus mechanical ones.
  5. Price the cache-write rate on your traffic. Opus cache writes bill at $6.25 per million, above the input rate.

Cache writes above the input rate, and what that means

Illustrative example

An Opus 4.8 workload sending a 6,000-token system prompt on every call, 200,000 calls a month. Rates from the price book: $5.00 input, $25.00 output, $0.50 cached read, and a five-minute cache write at $6.25 per million, derived at 1.25 times input.

No cache, prompt only: 200,000 x 6,000 = 1,200M tokens x $5.00 per million = $6,000
Cached read, prompt only: 1,200M x $0.50 per million = $600, a saving of $5,400
One cache write costs 6,000 tokens x $6.25 per million = $0.0375, against $0.03 for an uncached call
The write replaces an uncached call, so its extra cost is $0.0375 - $0.03 = $0.0075
Each later read saves $0.027, so break-even = $0.0075 / $0.027 = 0.28 reads
Rebuild the cache on every call and you pay $7,500, which is 25% worse than never caching

Caching takes $5,400 a month off this prompt and clears break-even on the first read. Only a cache rebuilt with no read at all loses, and then by 25%.

Every number, with its confidence and source

FigureWhat it meansConfidenceSource
$5.00 / $25.00 per millionClaude Opus 4.5, 4.6, 4.7, 4.8 and Opus 5 input and output rates, identical across all fivecalculatedCulpa price book rows for anthropic/claude-opus-4-5 through 4-8 and claude-opus-5, all $0.005 and $0.025 per 1k, verified against Anthropic pricing on 2026-08-01.
$6.25 per millionOpus five-minute cache write rate, above the input ratecalculated$0.005 per 1k input x 1.25, the published Anthropic cache-write multiplier, mirrored from the ledger's own derivation in migration 0020.

What a generic answer can’t know

A rate card can tell you five Opus versions cost the same. It can't tell you which version each of your services actually pinned, or that the oldest pin carries a prompt six hundred tokens longer than the rest. That needs per-call version and token data. Culpa records it on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.

Questions founders ask next

Is Claude Opus 4.8 more expensive than Opus 4.5?

No. Opus 4.5, 4.6, 4.7, 4.8 and the newer Opus 5 all bill $5.00 per million input and $25.00 per million output. Choosing between them is a behaviour decision, never a cost one.

What does an Opus cache write cost?

$6.25 per million for the five-minute window and $10.00 for the one-hour window, derived at 1.25 and 2 times the $5.00 input rate. Both sit above the input rate, so a cache you rebuild too often costs more than not caching.

Should I run a whole agent on Opus?

Rarely. Planning and synthesis benefit from it. Tool selection, routing and formatting usually run identically on a model costing a fifth as much, and splitting the loop by step type is generally the biggest saving available.

On your infrastructure

Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.


How Culpa works

Find the culprit. Not just the total.

Your dashboard shows what you spent. It stops short of who spent it. Culpa shows the conversation, the user and the feature behind it.

Your prompts stay local.

Culpa runs on your own infrastructure. What you send to a model reaches us at no point.

Every dollar has a name.

Follow any charge to the conversation, the user, the feature and the customer behind it.

See the bill before it lands.

Cost your next feature before you ship it. You get the likely bill and the worst case, at best, median, p90 and p99.

Three steps to your first answer.

1

Change one base URL.

Or drop in the Python or TypeScript library.

2

Find your most expensive conversation.

In the first session, not the first week.

3

Cost your next feature before you ship it.

Base URLhttp://localhost:4545/v1Your traffic keeps flowing if Culpa ever stops.

Why the bill went up

Example dashboard

Calls traced

418,209

across 3 projects

Spend this week

$378.41

+ $182 vs last week

Failed calls

312

74% retried, and you paid for all of them

+ $182 this week traced to one culprit

Spend over 14 days

$0$20$40$60$8024262830020406
user_384report_generatorconv_91fprompt_v1894,220 tokens3 retries$6.81

Most expensive users

user_384$38.42
user_119$21.07
user_562$14.90
user_204$8.30
user_871$5.10

Next week forecast

Best$180
Median$240
p90$310
p99$395

Graded against reality. Accuracy shown as results land.

Keep reading


Sources: Anthropic pricing. Last reviewed 2026-08-01, rates effective 2026-07-02. Plain text version.