Buy Crypto Markets Spot FuturesBTC Earn Event Centre Rewards Hub

You have a feature that needs an AI model behind it, a rough idea of how much text it will handle, and no idea what the bill looks like. The Claude API price is quoted per million tokens, and as ofYou have a feature that needs an AI model behind it, a rough idea of how much text it will handle, and no idea what the bill looks like. The Claude API price is quoted per million tokens, and as of

Learn/Market Insights/Others/How Much Do...laude Model

How Much Does the Claude API Cost? Price per Million Tokens for Every Claude Model

Beginner

Sep 16, 2026

NOT$0.0004268-6.05%

BILL$0.01255-5.85%

AI$0.0186-4.12%

You have a feature that needs an AI model behind it, a rough idea of how much text it will handle, and no idea what the bill looks like.

The Claude API price is quoted per million tokens, and as of September 2026 it runs from $1 to $10 on the input side depending on which model you pick.

Here is the full rate card, plus the three things people actually mean when they search for a Claude token price.

Key Takeaways

As of September 2026 the four Claude models you can buy directly are Haiku 4.5 at $1/$5, Sonnet 5 at $2/$10, Opus 5 at $5/$25, and Fable 5.1 at $10/$50 per million tokens.
Every Claude model charges output at five times its input rate, which makes response length the main driver of your bill.
Prompt caching cuts repeat input to a tenth of the input rate and the Batch API halves everything, making those two the only discounts that materially move the number.
A Claude subscription and an API key are separate meters, so paying for Pro or Max does not cover a single API token.
Anthropic has issued no cryptocurrency, so a Claude token price means a per-million-token API rate and any coin trading under the name is an unaffiliated third-party project.
Several retired versions still circulate in search results at their old prices, so check any version against Anthropic's own retirement schedule before budgeting from it.

What Is the Claude API Price per Million Tokens?

Claude API pricing is billed per million tokens, split between what you send the model and what it writes back.

As of September 2026 Anthropic lists four models any account can buy: Claude Haiku 4.5 at $1 input and $5 output, Claude Sonnet 5 at $2 and $10, Claude Opus 5 at $5 and $25, and Claude Fable 5.1 at $10 and $50.

Model	Input	Output	Cache hit	Batch input	Batch output
Claude Fable 5.1	$10	$50	$0.25	$5	$25
Claude Mythos 5.1 (limited availability)	$10	$50	$0.25	$5	$25
Claude Opus 5	$5	$25	$0.50	$2.50	$12.50
Claude Sonnet 5	$2	$10	$0.20	$1	$5
Claude Haiku 4.5	$1	$5	$0.10	$0.50	$2.50

All figures are dollars per million tokens, from Anthropic's published rate card as observed in September 2026.

Trust that page over any secondary source, since the rate card has changed twice this year.

How to Read a Rate Card: Input, Output, and Cache Columns

MTok means one million tokens, the unit every Claude price is quoted in.

Input covers everything you send: the user's message, your system prompt, the conversation history you replay, and any tool definitions.

Output covers only what the model generates.

Output costs five times input on every model, which is why a chatbot writing long answers costs more than a classifier returning one word.

Which Claude Models Can You Actually Buy Today?

Four models are available to any account with an API key: Claude Haiku 4.5, Claude Sonnet 5, Claude Opus 5, and Claude Fable 5.1.

A fifth, Claude Mythos 5.1, appears on the same rate card at the same $10 and $50, but sits behind an approval program rather than self-serve signup.

Fable and Mythos share one underlying model and differ in the safeguards applied to it, which is why their prices match.

Mythos runs through Anthropic's limited-availability program, so for most teams the top of the lineup is Fable 5.1.

Opus vs Sonnet vs Haiku vs Fable: The Tenfold Price Gap

The gap between Haiku 4.5 and Fable 5.1 is tenfold on input and tenfold on output.

Four things decide which one you pay for, and cost per token is only the first: the context window you need, whether a wrong answer costs more than a retry, and how much the task gains from multi-step reasoning.

Those are the dimensions because they change the answer. Benchmark scores rarely do.

Haiku 4.5 suits classification, extraction, and tagging at volume, where the job is narrow and a retry is cheap. Sonnet 5 is the sensible production default, and it now costs less than the Sonnet 4.5 and 4.6 releases that came before it.

Opus 5 earns its rate on multi-step reasoning where an error propagates into work you redo, and Anthropic positions Fable 5.1 for agentic coding and long-horizon problem solving.

The argument for paying more: a model that finishes on the first attempt can cost less per completed job than a cheap model you call three times.

Cost per token and cost per outcome are different numbers, and only one appears on a rate card.

Which Claude Version Numbers Are Current, and Which Are Retired?

Anthropic keeps several older versions callable at their listed prices and retires others on a published schedule.

Search results do not track that schedule, which is how a retired model's old price ends up in a budget.

Version	Status	Input	Output
Opus 5, 4.8, 4.7, 4.6, 4.5	Active	$5	$25
Opus 4.1 and Opus 4	Retired on the first-party API	$15	$75
Sonnet 5	Active	$2	$10
Sonnet 4.6 and 4.5	Active	$3	$15
Sonnet 4	Retired on the first-party API	$3	$15
Haiku 4.5	Active	$1	$5
Haiku 3.5	Retired on the first-party API	$0.80	$4

Sonnet 3.5 and Sonnet 3.7 no longer appear on the rate card, so a quoted Claude 3.5 Sonnet price is history rather than something you can buy.

Two traps are worth naming.

First, newer is not automatically pricier. Sonnet 5 undercuts the Sonnet releases before it, and the increase to $3 and $15 that Anthropic had scheduled for September 2026 was cancelled, with the introductory rate made permanent.

Second, sticker price and job cost can move apart. Anthropic notes that Claude 4.7 and later models use a newer tokenizer producing roughly 30% more tokens for the same text, so an identical document costs more on those models at an identical per-token rate.

Check any older version against Anthropic's deprecation and retirement schedule before you build a budget on it.

Why Is "Claude Token Price" Not a Cryptocurrency Price?

A Claude token price is the per-million-token rate you pay to send text through the API.

Anthropic has not issued a cryptocurrency, a coin, or an airdrop, so no Claude token carries any claim on the company or its models.

In this context a token is a unit of text, roughly four characters, not a tradable asset.

Tokens using the Claude or Anthropic name do appear on public crypto trackers, and they are unaffiliated with the company.

Anthropic itself is still private, a separate question covered in our guide to whether Anthropic stock is public yet.

What Anthropic Says About Unauthorized Offers

Anthropic publishes a warning of its own, and it is unusually blunt.

On its investor warning page the company states that any sale or transfer of its stock, or of any interest in its stock, that its board has not approved is void and will not be recognized on its books.

It does not permit special purpose vehicles to acquire its stock.

On public offers it goes further, saying third parties selling its shares to the general public through direct sales, forward contracts, or tokenized securities are likely either committing fraud or selling something that may have no value.

Its guidance to anyone pitched indirect access: assume the arrangement is invalid.

The company also lists what a scam pitch tends to look like:

Unsolicited contact by email, social media, or a messaging app
Claims of exclusive or time-limited access to shares
Requests for payment in cryptocurrency or by wire transfer
Pressure to commit before you can check anything
A claim that the deal is structured to get around the company's transfer restrictions
No documentation of board approval

What Does Exist: A Futures Contract, Not a Token

MEXC lists an Anthropic pre-IPO futures contract, and describing it accurately matters more than describing it attractively.

It is a USDT-margined perpetual contract, and it conveys no shares, no voting rights, no dividends, and no claim on the company.

Its quote is not a share price you could find on an exchange, because Anthropic has no public share price. MEXC describes a pre-IPO contract as tracking the expected share price of a company that has not yet gone public, so the number moves with expectations rather than with a live quote.

A product claiming to hand you shares runs into Anthropic's transfer restrictions. A cash-settled contract makes no ownership claim in the first place.

If Anthropic does list, MEXC documents what changes when a pre-IPO contract converts to standard stock futures.

How Do You Calculate What a Claude API Job Will Cost?

Multiply input tokens by the input rate, multiply output tokens by the output rate, divide each by a million, then add the two.

Anthropic puts one token at roughly four characters or 0.75 words of English, which makes a million tokens about 750,000 words.

Work a real job through it. A 10,000-word document summarized to 1,000 words is about 13,000 input tokens and 1,300 output tokens.

On Opus 5 that is 13,000 x $5 / 1,000,000 for input, or $0.065, plus 1,300 x $25 / 1,000,000 for output, or $0.033. The document costs about ten cents.

On Haiku 4.5, at a fifth of those rates, the same job costs about two cents.

Scale matters more than the unit price. Anthropic's own worked example puts 10,000 support conversations averaging roughly 3,700 tokens each at about $37 on Haiku 4.5.

Estimates get you a budget, not an invoice. For exact counts Anthropic provides a free token-counting endpoint, which matters because a token count is model-specific and an estimate from another tool will not match your bill.

Is the Claude Subscription Price the Same as the API Price?

No, and this catches people out constantly.

A Claude subscription buys access to the Claude apps at a flat monthly rate, while an API key bills per token with no monthly fee.

They are separate meters, and paying for one does not cover the other.

As of September 2026 Claude Pro runs $20 per month, with the Max plans at $100 and $200 for higher usage ceilings.

Team and Enterprise plans are priced per seat, with Enterprise billing API usage separately on top of the seat cost.

Plan structures change more often than API rates, so confirm current tiers on Anthropic's pricing page before committing.

Which costs less depends on the shape of your usage, not its size: steady all-day interactive work suits a subscription, bursty programmatic traffic suits per-token billing.

What Does Claude Code Cost?

Claude Code has no price of its own.

It comes included in the paid Claude plans, or it bills per token against an API key at the same model rates as any other request.

On API billing the cost tracks how much context each session pushes through, which is why a session spanning a whole feature costs more than one fixing a typo.

Its usage command shows the session's token count and an estimated dollar figure, which Anthropic's own documentation flags as an estimate rather than your actual bill.

Do Third-Party Platforms Charge a Different Claude Price?

Third-party channels generally match Anthropic's per-token rates rather than undercut them, then add a fee of their own elsewhere.

OpenRouter lists Claude models at the same per-million-token rates Anthropic charges, and takes a percentage on credit purchases instead of marking up tokens.

Cloud platforms do the same differently: regional and multi-region endpoints carry a 10% premium over global ones, buying a guarantee about where inference runs rather than extra capability.

On the first-party API, pinning inference to the United States adds a 1.1x multiplier across every token category on Claude 4.6 and later models.

The channels worth checking all match Anthropic on raw tokens rather than undercutting it. A price gap you find is almost always a fee or a routing guarantee, not a discount.

How Can You Lower Your Claude API Bill?

Four levers do real work, and they are not equal.

Prompt caching is the largest for most workloads. A cache hit costs 10% of the input rate, and 2.5% on Fable 5.1, so replaying a long system prompt gets dramatically cheaper.

A five-minute cache write costs 1.25x the input rate and pays for itself after a single read, while a one-hour write costs 2x and needs two reads to break even.

The Batch API takes 50% off both input and output for anything that can wait, which covers most overnight and backfill work.

Third, route extraction and classification to Haiku 4.5 and reserve the expensive models for requests that need them.

The fourth is knowing what you do not have to do. Claude 4.6 and later models include the full one-million-token context window at standard pricing, so a 900,000-token request bills at the same per-token rate as a 9,000-token one and aggressive chunking buys you nothing.

One cost sits outside the token model: web search bills $10 per 1,000 searches on top of tokens.

FAQ

How much does the Claude API cost?

As of September 2026 it runs from $1 input and $5 output per million tokens on Haiku 4.5 up to $10 and $50 on Fable 5.1.

What is Claude token pricing per million tokens?

Claude prices input, cache reads, cache writes, and output separately, with standard rates spanning $1/$5 to $10/$50 across the current lineup.

Is the Claude API cheaper than a Pro or Max subscription?

For light or bursty programmatic use the API usually costs less, while steady all-day interactive use generally favors a subscription.

How much does Claude Opus 5 cost?

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens, or half that through the Batch API.

How much does Claude Fable 5.1 cost per token?

Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens, with cache hits at $0.25.

How much does Claude Code cost?

Nothing beyond a paid Claude plan that includes it, or standard per-token API rates if you run it on an API key.

Is there a Claude crypto token?

Not from Anthropic, which has issued no cryptocurrency, so any token trading under the Claude name is a third-party project unaffiliated with the company.

What is the cheapest way to use Claude?

Route simple work to Haiku 4.5, turn on prompt caching, and send anything that can wait through the Batch API at half price.

Do Claude Opus 4.7 and 4.8 still work?

Yes, both remain active on the API at $5 and $25 per million tokens, alongside Opus 4.5, 4.6, and Opus 5.

Conclusion

Rate cards change, and Anthropic cancelled a scheduled increase on this one in September.

The structure holds steady: output costs five times input, caching and batching are the discounts that matter, and the cheapest model that finishes your job wins.

Check the figures against Anthropic's own page before committing real budget. If it is Anthropic the company you want exposure to rather than its API, start with how its valuation and funding rounds stack up.

Market Opportunity

Notcoin Price(NOT)

$0.0004268

$0.0004268$0.0004268

-4.51%

USD

Notcoin (NOT) Live Price Chart

This article is provided by MEXC for informational purposes only and does not constitute financial or investment advice. Cryptocurrency markets involve significant risk. Please conduct independent research or consult a qualified professional before making any investment decisions. The views expressed do not necessarily represent those of MEXC or its affiliates.