Claude API Pricing 2026: Every Model per Million Tokens, Cache Read and Write Prices, and Cost per Conversation

Claude API Pricing in 2026: Every Model per Million Tokens, Cache Prices, and What Pro and Max Cost

Published 26 September 2026 · Updated 27 October 2026 · Every price below was read from Anthropic's pricing documentation, the Claude plans page and the Claude Help Center on 27 October 2026.

The Claude API costs between $0.10 and $10 per million input tokens and between $0.50 and $50 per million output tokens on Anthropic's four current models, as of 27 October 2026: Claude Haiku 5.5 at $0.10 and $0.50 for prompts up to 100,000 tokens (and $0.50 and $2.50 above that), Claude Sonnet 5.5 at $2 and $10, Claude Opus 5.5 at $4 and $20, and Claude Fable 5.1 at $10 and $50. Cache reads cost between $0.01 and $0.25 per million tokens, the Batch API halves every rate, and Haiku 5.5 is the only current model whose price depends on prompt length. Claude's apps are priced separately: Free is $0, Pro is $20 a month ($17 a month billed annually), and Max is $100 or $200 a month, which since 7 October 2026 includes $100 or $200 a month in API credits.

Those rates are the easy part. What decides a Claude bill is how many tokens each request carries, and in a multi-turn product that number grows with every turn, because the whole conversation is sent again each time. The worked example further down shows a 40-turn chat on Sonnet 5.5 costing $1.09 without prompt caching and $0.24 with it, and the same chat on Haiku 5.5 costing under six cents without caching.

Claude API price per million tokens, current models

All figures are US dollars per million tokens, as of 27 October 2026, from Anthropic's pricing documentation and the Claude pricing page. Claude Mythos 5.1, offered to Project Glasswing participants, carries the same rates as Fable 5.1. All four current models have a 1 million token context window and generate up to 128,000 output tokens per request, according to the models overview.

Six changes since this page first went up on 26 September explain why most pricing pages are now out of date. The dates come from the Claude Platform release notes:

On price alone the current line-up is simple to rank. Sonnet 5.5 costs half of Opus 5.5 per token, Fable 5.1 costs two and a half times Opus 5.5, and Haiku 5.5 under 100,000 tokens costs one twentieth of Sonnet 5.5 and one tenth of Haiku 4.5. Which model is good enough for a given job is a quality question the rate card cannot answer, so run your own prompts through two adjacent tiers before committing a workload to either.

Haiku 5.5's 100K-token tier: what crosses the line and what it costs

Claude Haiku 5.5 charges five times more per token on any request whose prompt is over 100,000 tokens, and that higher rate applies to the whole request, output included. Anthropic's pricing documentation spells out three rules that decide which side of the line a request falls on:

• The prompt length counts all of the request's input tokens, including cache reads and cache writes, so a 95,000-token cached history plus a 6,000-token new message is a 101,000-token prompt.
• Each request is priced on its own. A request over the threshold pays the higher prices even when most of its prompt is a cache hit, and earlier requests keep the prices they were billed at.
• The 1.1 times US-only inference multiplier applies on top of the higher prices.

The jump is easiest to see on a single conversation. Using the chat model from the worked example below (a 2,000-token system prompt, 150-token user messages and 350-token replies), turn 196 sends 99,650 tokens and turn 197 sends 100,150. Without caching, turn 196 costs $0.0101 and turn 197 costs $0.0510, five times as much for 500 more tokens. With a warm cache the same two turns cost $0.0012 and $0.0062. Every turn after that stays in the higher tier for as long as the full history is sent.

Two things keep this in proportion. Haiku 5.5 above 100,000 tokens is still a quarter of Sonnet 5.5's price per token ($0.50 against $2 for input), so crossing the line does not make Sonnet the cheaper model. And Anthropic's launch post says prompts up to 100,000 tokens made up about 90% of requests to the previous Haiku model, so most workloads never reach the higher tier. The ones that do are long chats, agents that re-send a growing tool history, and retrieval pipelines that put large documents into every call. For those, compacting the history before it reaches 100,000 tokens, or sending a bounded context instead of the full transcript, keeps every request in the cheaper tier.

Older models still on the price sheet

Anthropic's pricing page still lists earlier models, and several cost more than their replacements. The status column comes from Anthropic's model deprecations page; a retired model fails every request on the Claude API.

The prompt caching documentation notes that Opus 4.1 and Sonnet 4 remain available on Amazon Bedrock and Google Cloud after their Claude API retirement. At Anthropic's list rates, a workload still on Opus 4.1 pays $15 and $75; the same tokens on Opus 5.5 cost 73% less. A workload on Sonnet 4.6 pays $3 and $15, so moving it to Sonnet 5.5 cuts the token bill by a third, although a model change is a quality decision as well as a price one and deserves the same evaluation as any other.

Sonnet 5 or Sonnet 5.5? Both cost $2 and $10, and Sonnet 5.5 reads its cache at half the price ($0.10 against $0.20). Anthropic's launch post says Sonnet 5.5 generates output 30% or more faster and, in Anthropic's own testing, costs up to 30% less per task because it needs fewer tokens for the same work. The release notes list five ways code written for Sonnet 5 can break on Sonnet 5.5: thinking: {"type": "disabled"} has to become {"type": "between_tools"}, forced tool use (tool_choice of any or tool) returns a 400 error, thinking blocks are tied to the model and conversation, the older computer_20251124 tool is not accepted on the Claude API and Google Cloud, and the advisor tool rejects Opus 4.8, Opus 4.7 and Sonnet 5 as advisors. On a cache-heavy workload the switch is worth those fixes.

What one request actually bills

Every request is billed as input tokens plus output tokens, and input is more than the user's latest message. Anthropic's context window documentation lists what counts: the system prompt, every message in the messages array (including tool results, images and documents), the tool definitions, and the output Claude generates. In a conversation, "previous turns are preserved completely", which means turn 40 pays again for turns 1 to 39.

Thinking tokens are billed as output. The extended thinking documentation reports them in usage.output_tokens_details.thinking_tokens as part of the billed output, so a reasoning-heavy request can cost several times what its visible answer suggests. Claude Opus 4.5 and every model numbered 4.6 or higher, which covers all four current models, keep earlier turns' thinking blocks in context and bill them as input; Sonnet 4.5, Haiku 4.5 and earlier models strip them.

The tokenizer matters too. Anthropic's pricing page says Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text, with the exact increase depending on content. A per-token price comparison between Haiku 4.5 and Haiku 5.5, or between Sonnet 4.6 and Sonnet 5.5, therefore overstates the saving on a fixed piece of text, so count tokens with the token counting endpoint and the target model's ID before you budget, and read the usage fields on every response after.

Tools add input you never wrote. Whenever a request includes tools, the API adds a tool-use system prompt: 286 tokens on Opus 5.5, Sonnet 5.5 and Haiku 5.5 with tool_choice set to auto or none, 354 on Sonnet 5 and 675 on Opus 4.7, on top of the tool definitions themselves, according to the pricing page's tool-use table.

When prompt caching pays for itself

Prompt caching stores a prompt prefix so that later requests read it instead of paying full input price for it. The prompt caching documentation sets the multipliers against the base input rate: a 5-minute cache write costs 1.25 times, a 1-hour cache write costs 2 times, and a cache read costs 0.1 times on most models. Opus 5.5 and Sonnet 5.5 read at 0.05 times, and Fable 5.1 and Mythos 5.1 at 0.025 times, which is why Sonnet 5.5's cache read ($0.10) costs the same as Haiku 4.5's.

The break-even follows directly. Two requests sharing a 5-minute cached prefix pay 1.35 times the prefix instead of 2 times (1.3 times on Opus 5.5 and Sonnet 5.5), so caching saves money from the first reuse. A 1-hour write pays 2.1 times across two requests against 2 times uncached, so it loses on one reuse and wins from the second. The 1-hour cache is for traffic that returns after more than five minutes but less than an hour, such as a user who pauses between messages; the 5-minute cache suits rapid turns and agent loops.

Four details decide whether caching works in practice:

• Minimum length. A prefix shorter than the model's minimum is not cached, and the API returns no error. The minimum is 512 tokens on Fable 5.1, Mythos 5.1, Opus 5.5, Opus 5, Sonnet 5.5, Fable 5, Mythos 5 and Haiku 5.5; 1,024 tokens on Opus 4.8, Sonnet 5 and Sonnet 4.6; 2,048 tokens on Opus 4.7; and 4,096 tokens on Opus 4.6, Opus 4.5 and Haiku 4.5.
• Order of invalidation. The cache follows the order tools, then system, then messages, and a change at one level invalidates that level and everything after it. Changing a tool definition invalidates everything; changing tool_choice or adding or removing an image invalidates only the messages.
• Breakpoints. A request can carry up to four cache breakpoints, and automatic caching (a single top-level cache_control field) uses one of them and moves forward as the conversation grows.
• Rate limits. Cache hits are not deducted from your input-tokens-per-minute limit on most models; only uncached input and cache writes are.

Check cache_read_input_tokens and cache_creation_input_tokens in every response. A cache that silently misses costs ten times the read price on every request (twenty times on Opus 5.5 and Sonnet 5.5, forty on Fable 5.1), and the only place it shows up is the usage object.

Batch API, long context and the smaller line items

Batch API. Asynchronous batch requests cost 50% of the standard input and output rates on every model, as the first table shows, and the discount stacks with caching. Anything that does not need an answer within the conversation, such as nightly summarisation, evaluation runs or back-filling a data set, belongs there.

Long context. Claude 4.6 and later models, except Haiku 5.5, bill their full 1 million token context window at standard rates: a 900,000-token request pays the same per-token price as a 9,000-token one. Haiku 4.5 and Sonnet 4.5 have 200,000-token windows.

Fast mode. Opus 5.5 runs at $8 and $40 per million tokens in fast mode, twice its standard rate; Opus 5 and Opus 4.8 run at $10 and $50 in a research preview. Fast mode is not available with the Batch API, and caching and data-residency multipliers apply on top.

US-only inference. Setting inference_geo to "us" applies a 1.1 times multiplier to every token category on Claude 4.6 and later models. Earlier models reject the parameter with a 400 error.

Tools and runtime. Web search costs $10 per 1,000 searches plus the tokens it adds, and web fetch costs only its tokens. Code execution is free when used alongside the web_search_20260209 or web_fetch_20260209 tools (or later versions); otherwise it bills $0.05 per container-hour beyond 1,550 free hours per organisation per month, with a 5-minute minimum per execution. Claude Managed Agents sessions cost $0.08 per session-hour while running, plus tokens.

Cloud marketplaces. Amazon Bedrock and Claude Models on Google Cloud (Google managed) set and invoice their own Claude prices, and their regional and multi-region endpoints carry a 10% premium over global endpoints for Claude 4.5 models and later. Claude Platform on AWS and Claude in Microsoft Foundry rate usage at Anthropic's standard prices and bill it as Claude Consumption Units at $0.01 each.

Usage tiers. The rate limits page caps monthly spend at $500 on the Start tier, $1,000 on Build and $200,000 on Scale; organisations move up automatically with usage history, and the Custom tier has no cap.

What a real conversation costs

Take a chat assistant with a 2,000-token system prompt, where each user message is 150 tokens and each reply is 350 tokens. The first turn sends 2,150 tokens. By turn 40 a single request carries 21,650 tokens, because it includes the 39 exchanges before it, and the whole 40-turn conversation has consumed 476,000 input tokens to produce 14,000 output tokens.

Prices as of 27 October 2026. "Cache warm" assumes automatic caching with a 5-minute cache and replies arriving inside that window, so each turn writes only the new exchange and reads everything before it. Every model is given the same token counts; because Claude 4.7 and later models tokenize the same text into more tokens, real counts on the newer models will run somewhat higher. The 100-turn conversation peaks at 51,650 tokens per request, so Haiku 5.5 stays in its lower tier throughout.

Three things stand out. Input tokens, not output, make up most of the bill once a conversation passes a few turns. The cost per conversation grows faster than the number of turns: going from 40 to 100 turns multiplies turns by 2.5 and the uncached bill by more than five, because each turn carries every turn before it. And Sonnet 5.5's cheaper cache read is worth real money on long conversations: at 100 turns with a warm cache it costs $0.74 against Sonnet 5's $1.01, for the same input and output prices.

At 10,000 such 40-turn conversations a month on Sonnet 5.5, that is roughly $10,920 without caching and $2,396 with it; on Haiku 5.5 it is roughly $546 and $142. To estimate your own bill, multiply each token category by its rate (uncached input, 5-minute and 1-hour cache writes, cache reads, and output including thinking), sum across the turns of a typical session, and multiply by sessions per month; on Haiku 5.5, price each request whose prompt passes 100,000 tokens at the higher row.

Caching is the first lever. It stops working when users leave and come back after the cache has expired, and the history has to be written again at the higher write price. We build Maximem Synap, a memory layer that sends a bounded context block (2,000 tokens by default) instead of the full transcript, so this cost is the problem we work on every day, and on Haiku 5.5 a bounded context also keeps every request under the 100,000-token line. The arithmetic for when that beats caching, and when it does not, is in how to reduce LLM token costs in long conversations, and the context-size side is in our guide to LLM context window sizes in 2026.

How much Claude costs per month: the plans

Claude's apps (web, desktop, mobile, Claude Code and the Office and browser extensions) are sold as subscriptions, priced per person, from the Claude plans page as of 27 October 2026:

Team plans cover 2 to 150 seats, seat types can be mixed, and a team's pooled API credits are capped at $500 a month. Annual billing saves $40 a year on Pro and $60 a seat on Team Standard. Enterprise adds SCIM, audit logs, role-based access, custom data retention, IP allowlisting, spend controls and a HIPAA-ready option on top of Team.

Free vs Pro vs Max: what actually changes

The three individual plans differ mainly in how much you can use, and Pro is where the developer tools start.

Free covers chat on web, desktop and mobile with web search, file creation and code execution, memory, Artifacts, Claude Design, Slides and Docs, and up to five projects, on Sonnet and Haiku models only.

Pro adds more usage, Claude Code, Claude Science, Claude Dashboards (in beta), projects, the Opus models, Claude in Chrome and Microsoft 365, and task hand-off and scheduling.

Max keeps every Pro feature and raises the allowance to five or twenty times Pro's per-session usage, with higher output limits, early access to new models, priority access when traffic is high, and the monthly API credits. The Max plan article is explicit that the two Max tiers differ in usage, not in models.

Anthropic does not publish message counts for any plan. On Pro and Max, the session allowance resets every five hours, and a separate weekly limit applies across all models and resets at a fixed time assigned to your account. Chat and Claude Code draw from the same allowance, so a long coding session reduces what is left for chat.

The practical rule: stay on Free until you hit its limits regularly, move to Pro when you want Claude Code or the Opus models, and move to Max when you run out of Pro's weekly allowance often enough that the extra $80 or $180 a month is cheaper than waiting. A developer who also spends on the API should count Max's credits against that difference.

Subscription or API?

The Pro plan does not include API usage; the Pro plan article says paying for the API needs separate Console access. Max and Team are different since 7 October 2026: they include monthly API credits, which the API credits documentation describes in detail.

• Amounts. $100 a month on Max 5x and $200 on Max 20x. Team plans get $20 per Standard seat and $100 per Premium seat, pooled into one balance capped at $500 a month.
• What they pay for. The Claude API, Claude Managed Agents, the Claude Agent SDK and the Console Playground, and only through the Claude Console, not through Bedrock, Google Cloud or Foundry.
• What they do not pay for. Claude Code and extra usage in the Claude apps. Credits also do not change your usage limits in either.
• Rules. Unused credits expire at the end of each billing cycle and do not roll over. New subscribers can claim them after 7 days, from Settings, then Billing, by linking one Console organisation, which you cannot change yourself afterwards. Free, Pro and Enterprise plans are not eligible.

Two billing surprises remain:

• Claude Code and the API key. If ANTHROPIC_API_KEY is set in your environment, Claude Code authenticates with it instead of your subscription and bills your API account at token rates, even while you are paying for Max. Unset the variable if you meant to use the plan.
• The Agent SDK. Signed in with your Claude plan, the Agent SDK and claude -p draw from your subscription limits, as they have since Anthropic put its planned change on hold on 15 June 2026. They draw on the new API credits only when you use an API key from the linked Console organisation, according to the Agent SDK help article.

For a team building a product, the question mostly answers itself. Plans are priced per person for people using Claude's own apps; a product that serves your users runs on the API, billed per token, and the useful comparison is between API configurations (model, caching, batch) rather than between an API bill and a subscription. For an individual developer, Pro or Max is usually the cheaper way to use Claude Code heavily, Max now also covers a small prototype's API bill, and the API is the right way to run anything automated in production.

Frequently asked questions

How much does the Claude API cost per million tokens?

As of 27 October 2026, Claude Haiku 5.5 costs $0.10 input and $0.50 output for prompts up to 100,000 tokens and $0.50 and $2.50 above that, Claude Sonnet 5.5 costs $2 and $10, Claude Opus 5.5 costs $4 and $20, and Claude Fable 5.1 costs $10 and $50. Cache reads cost $0.01 to $0.25 per million tokens, and the Batch API halves the input and output rates.

What are Claude's cache read and cache write prices?

A 5-minute cache write costs 1.25 times the input price and a 1-hour write costs 2 times: $2.50 and $4 on Sonnet 5.5, $5 and $8 on Opus 5.5, $12.50 and $20 on Fable 5.1, and $0.125 and $0.20 on Haiku 5.5 under 100,000 tokens. Cache reads cost $0.10 on Sonnet 5.5, $0.20 on Opus 5.5, $0.25 on Fable 5.1 and $0.01 on Haiku 5.5 under 100,000 tokens ($0.05 above).

Why does Claude Haiku 5.5 cost more above 100,000 tokens?

Anthropic prices Haiku 5.5 by prompt length. A request whose prompt, counting cache reads and writes, is over 100,000 tokens pays $0.50 input and $2.50 output instead of $0.10 and $0.50, five times more, on that whole request. Requests under the line are unaffected, and each request is priced on its own.

How much does Claude Opus 5.5 cost?

$4 per million input tokens and $20 per million output tokens, with 5-minute cache writes at $5, 1-hour writes at $8 and cache reads at $0.20. Batch requests cost $2 and $10, and fast mode costs $8 and $40.

Should I use Claude Sonnet 5 or Sonnet 5.5?

Both cost $2 input and $10 output, but Sonnet 5.5 reads its cache at $0.10 instead of $0.20, and Anthropic reports it is 30% or more faster and up to 30% cheaper per task in its testing. Check the five breaking changes in the release notes, such as forced tool use returning a 400 error, before you switch.

How much does Claude cost per month?

Free is $0, Pro is $20 a month or $17 a month billed annually, Max is $100 (five times Pro usage) or $200 (twenty times), Team is $25 a seat a month or $20 billed annually, with Premium seats at $125 or $100, and Enterprise is $20 a seat a month billed annually plus usage at API rates.

Is API usage included in Claude Pro or Max?

Pro includes no API usage. Since 7 October 2026, Max includes $100 or $200 a month in API credits and Team includes up to $500 a month in pooled credits, usable on the Claude API, Managed Agents and the Agent SDK through the Claude Console but not in Claude Code. Claude Code on a Pro or Max plan uses the plan's allowance unless an ANTHROPIC_API_KEY environment variable is set, in which case it bills the API.

Is there a free Claude API tier?

Anthropic's pricing pages list none. The Free plan covers the Claude apps, not the API, and the API credits that come with Max and Team are part of paid plans.

What is the cheapest way to use Claude?

For personal use, the Free plan. For API workloads, Claude Haiku 5.5 through the Batch API at $0.05 input and $0.25 output per million tokens on prompts up to 100,000 tokens, with prompt caching on any repeated prefix.

Does long context cost more on Claude?

Only on Haiku 5.5, which charges its higher tier above 100,000 tokens. Sonnet 5.5, Opus 5.5, Fable 5.1 and the other Claude 4.6 and later models bill their full 1 million token window at standard rates. The cost of long context still comes from volume: every token in the window is billed on every request that carries it.

How do I estimate my monthly Claude API bill?

Multiply tokens by rate for each part of a request: uncached input, cache writes, cache reads and output, including thinking tokens. For conversations, remember that input per turn grows with the history, as the worked example above shows. Count tokens with the token counting endpoint before launch, then read the usage fields on live traffic, which is where a cache that is not hitting will show up.

What this means for your bill

The rate card sets the price of a token, and your architecture sets how many tokens you buy. Prompt caching and the Batch API are the two largest discounts Anthropic offers, and both are configuration rather than engineering. Haiku 5.5's 100,000-token line is the first Claude price that punishes a long prompt directly instead of only through its volume, so how much of each conversation you choose to send again on every turn is now a design decision with a price threshold attached: products that remember a user in a bounded context pay the low tier on every request, and products that replay the whole history eventually pay five times as much per token for it.

---

Sources, all retrieved 27 October 2026: Anthropic pricing documentation; Claude plans and pricing; models overview; model deprecations; Claude Platform release notes; API credits for Max and Team plans; rate limits; Claude Platform documentation on prompt caching, extended thinking, context windows and token counting; Claude Help Center articles on the Pro plan, the Max plan, Claude Code on Pro and Max and the Agent SDK on paid plans; Anthropic's announcements of Claude Haiku 5.5 and Claude Sonnet 5.5. Worked examples are our own arithmetic on those rates.