What Does Claude Opus 4.8 Cost?
Claude Opus 4.8 is reported at $5.00 per million input tokens and $25.00 per million output tokens, unchanged from Opus 4.7. That number is reported rather than confirmed, because Anthropic does not publish an isolated price for Opus 4.8.
Opus 4.8 shipped on 28 May 2026 as an incremental Opus-tier update focused on reliability, and reporting on it has been consistent that it stayed in the same pricing tier as the model before it. What is missing is a vendor price line that names Opus 4.8 and puts a number beside it.
Where it does appear on a price page, it is grouped with other older models into a legacy tier quoted as a range rather than as a rate:
| Opus 4.8 pricing question | What is actually published |
|---|---|
| Legacy tier input range | $3 to $10 per million tokens |
| Legacy tier output range | $15 to $50 per million tokens |
| Isolated Opus 4.8 input price | Not published separately, reported at $5 |
| Isolated Opus 4.8 output price | Not published separately, reported at $25 |
| Batch, cache and residency rates for 4.8 | Not published separately |
| Per-plan consumer availability for 4.8 | Not published |
The reported $5 and $25 sit inside that published range, which is the strongest thing that can honestly be said for them. Plan on those numbers, but confirm them in your own console before committing a quarter of spend. AI Perks tracks the credit programs that cover bills like these: $7.7M across 194 companies.
The Opus Rate Anthropic Does Publish Precisely
Claude Opus 5, released 24 July 2026, is confirmed at $5.00 per million input tokens and $25.00 per million output tokens on Anthropic's own pricing page, along with its batch, cache and residency modifiers.
It is the one Opus rate you can cite without hedging. The confirmed card:
| Meter | Claude Opus 5 |
|---|---|
| Base input | $5.00 / MTok |
| Base output | $25.00 / MTok |
| Batch API input | $2.50 / MTok |
| Batch API output | $12.50 / MTok |
| Prompt cache read | $0.50 / MTok |
| US-only inference | 1.1x the standard rate |
Two things follow. The Opus headline rate has not moved across 4.7 and 5, which is why the reported 4.8 figure between them is credible without its own price line. And those modifiers are published for Opus 5, not 4.8, so any batch or cache saving you model for 4.8 is an inference rather than a quoted rate.
Three Opus releases landed inside about three months in 2026, and the headline rate told the same story each time:
| Release | Date | What changed | Headline rate |
|---|---|---|---|
| Claude Opus 4.7 | 16 April 2026 | Frontier update that powers Claude Code, new tokenizer | $5 / $25, confirmed, unchanged from 4.6 |
| Claude Opus 4.8 | 28 May 2026 | Incremental update focused on reliability | $5 / $25 reported, not published separately |
| Claude Opus 5 | 24 July 2026 | Near-Fable performance on many tasks, long-running agents | $5 / $25, confirmed |
The Opus tier has been price-stable across these releases, so moving inside the line is a capability decision rather than a budget one. What moved invoices over this window was not the rate card at all.
Why the Same $5/$25 Does Not Mean the Same Bill
Opus 4.7 shipped with a new tokenizer that Anthropic says can increase effective token counts by up to 35% for the same text. Opus 4.8 inherits it. The rate held, the billable token count did not.
This is the largest uncontested cost change in the recent Opus line and it never touched a price list. Billing meters tokens, not characters, so identical text can meter materially higher on 4.7 and 4.8 than it did before.
Applying Anthropic's own stated ceiling to the reported Opus-tier rate:
| Same prompts, same text | Metered volume | Est. monthly cost |
|---|---|---|
| Pre-4.7 tokenizer | 50M in / 10M out | $500 |
| Opus 4.7 and 4.8 tokenizer, at the 35% ceiling | up to 67.5M in / 13.5M out | up to $675 |
Up to 35% is a stated ceiling, not an average, and where your workload lands depends on your text. A team moving from a pre-4.7 model to Opus 4.8 should expect an identical rate card and a larger invoice for identical inputs. A team moving from 4.7 to 4.8 should expect no tokenizer change, because both sit on the newer one.
What an Opus-Tier Workload Costs Per Month
At 50 million input and 10 million output tokens a month, an Opus-tier workload costs an estimated $500 at the standard rate, $550 with US-only inference and $250 on the Batch API.
That volume stands in for a small team running a coding assistant in production. Every figure below is arithmetic on the rates above, and the batch, cache and residency rows use modifiers published for Opus 5, since none are published for 4.8.
| Configuration | Est. monthly cost at 50M in / 10M out |
|---|---|
| US-only inference, 1.1x multiplier | $550 |
| Standard rate | $500 |
| 80% of input served from cache reads | $320 |
| Batch API | $250 |
The cache row excludes cache write charges, which depend on how often your prefix changes. The spread between the top and bottom rows is $300 a month at one volume, at one rate. Configuration moves this bill more than the choice between 4.7, 4.8 and 5 does, because those three carry the same headline rate.
A modest credit grant covers a year of any row in that table. AI Perks exists to find you one.
What Anthropic Has Not Published About Opus 4.8
Several things people assume about Opus 4.8 are absent from the record, and this page will not invent them.
No isolated price. The $5 and $25 are reported and sit inside the published legacy range, but they are not quoted for this model on its own.
No retirement or deprecation date. There is no dated lifecycle commitment for Opus 4.8 in the record behind this page, so none is printed here. Dated schedules for legacy Claude models circulate widely, and are worth exactly as much as the source you can trace them to.
No free-tier allowance. Opus is not included in the Claude free plan and the Free and Pro plans default to Sonnet, but which plan serves Opus 4.8 specifically is not published per model.
No benchmark scores, context window, max output or knowledge cutoff figures. These get repeated confidently for Opus 4.8 and none are confirmed here. A wrong context window is an outage rather than a rounding error, so read the model reference yourself.
No xAI or Grok token prices. Public figures conflict badly enough that anything printed here would be a guess.
Free Credits That Cover an Opus-Tier Bill
A $500 a month Opus workload is an estimated $6,000 a year. Credit programs routinely cover several multiples of that.
AI Perks tracks $7.7M in credits across 194 companies, spanning model providers, compute and the rest of the stack a production AI product bills for.
Anthropic runs Claude for Startups as a general program rather than a model-specific one, so nothing about it is tied to Opus 4.8 or any single release. Reported grant sizes span a wide band, roughly $5,000 to $100,000, and that range is reported rather than confirmed and varies by company.
Two things matter before planning around credits. Model credits and compute credits are separate bills, so a team with Anthropic credits still pays its inference host, database and observability vendors. And programs differ in what they look for, which is why breadth beats betting a year of runway on one name.
The AI Model category at getaiperks.com is where these sit, alongside the compute and tooling programs that land on the same invoice.
Frequently Asked Questions
How much does Claude Opus 4.8 cost per million tokens?
Opus 4.8 is reported at $5.00 per million input tokens and $25.00 per million output tokens, unchanged from Opus 4.7, though Anthropic does not publish that rate for the model on its own. It appears in a legacy tier quoted as $3 to $10 input and $15 to $50 output, and the reported figures sit inside that band. Credit programs are listed at getaiperks.com.
Is the $5/$25 figure for Opus 4.8 confirmed?
No. It is reported and internally consistent, not confirmed by an isolated vendor price line. The same rate is confirmed for Opus 4.7 before it and Opus 5 after it, which is what makes it a reasonable planning number for the model in between.
When will Claude Opus 4.8 be retired?
No retirement or deprecation date for Opus 4.8 is confirmed in the record behind this page, so none is stated here. Legacy is a pricing and update status, not a dated schedule. Check the vendor's own lifecycle page before pinning a model for a year.
Does Claude Opus 4.8 have a free tier?
The Opus tier is not included in the Claude free plan, and the Free and Pro plans default to Sonnet. Per-model plan availability for Opus 4.8 is not published, so treat any exact free allowance quoted for it as unverified.
Why did my Claude bill rise when the Opus rate did not change?
Most likely the tokenizer. Opus 4.7 introduced one that Anthropic says can raise effective token counts by up to 35% for the same text, and Opus 4.8 uses it. Identical prompts meter more tokens, so an unchanged $5 and $25 still produces a larger invoice.
Should I use Opus 4.8 or Opus 5?
Opus 5, released 24 July 2026, carries a confirmed $5 and $25 rate, published batch and cache modifiers, and Anthropic's positioning as performance close to Fable on many tasks at roughly half the price for long-running agentic work. That makes it the cleaner default for new work. Credits cover either at getaiperks.com.
Every confirmed rate on this page is labelled confirmed, and every reported one is labelled reported. Now go get someone else to pay it.