Claude Sonnet 5.5: price and what it's for

Claude Sonnet 5.5 price, in short

Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens on the Claude API, as of October 27, 2026. The Batch API halves both rates. A prompt cache read costs only $0.10 per million tokens. In the Claude apps, the pricing page lists Sonnet on the Free, Pro, Max, Ultra and Team plans, so you do not need the API to use it.

MTok on this page means one million tokens. A token is a small piece of text, often part of a word. To get a feel for how much text that is, read how much one million Claude tokens buys first.

Every API price here comes from Anthropic's pricing documentation and the Sonnet 5.5 model page, checked on October 27, 2026. Prices change. Check those two pages again before you commit to a budget.

What is Claude Sonnet 5.5?

Sonnet 5.5 is the middle model of Anthropic's 5.5 family. It was released on September 28, 2026, six days after Opus 5.5 arrived on September 22. Haiku 5.5 followed on October 7. Anthropic's release notes describe it as "a faster, lower-cost complement to Claude Opus 5.5".

The facts you need to call it from code:

• Model ID on the Claude API: claude-sonnet-5-5. On Amazon Bedrock the ID is anthropic.claude-sonnet-5-5, so do not copy one into the other.
• Context window: 1M tokens. Maximum output: 128K tokens per request on the standard Messages API, or up to 300K on the Message Batches API with the output-300k-2026-03-24 beta header.
• Reliable knowledge cutoff: June 2026.
• Adaptive thinking is on by default, at effort high.

The model is on the Claude API for all customers. It is also on Amazon Bedrock, Claude Platform on AWS, Claude Models on Google Cloud and Microsoft Foundry. Anthropic says it will not retire the model before September 28, 2027.

Sonnet 5.5 API price: every rate on one chart

[
  {
    "label": "Base input",
    "usd_per_mtok": 2
  },
  {
    "label": "5-minute cache write",
    "usd_per_mtok": "2.50"
  },
  {
    "label": "1-hour cache write",
    "usd_per_mtok": 4
  },
  {
    "label": "Cache read",
    "usd_per_mtok": "0.10"
  },
  {
    "label": "Output",
    "usd_per_mtok": 10
  },
  {
    "label": "Batch input",
    "usd_per_mtok": 1
  },
  {
    "label": "Batch output",
    "usd_per_mtok": 5
  }
]

The base rates are $2 in and $10 out. Output costs five times as much as input. This means a task that writes a lot of code costs more than a task that only reads a lot of code.

Sonnet 5.5 has no long-context premium. Anthropic's pricing page says a 900K-token request is billed at the same per-token rate as a 9K-token request, anywhere inside the 1M-token window. Haiku 5.5 is the one current model that charges more for long prompts, which matters when you compare the two.

One setting raises every rate. If you set inference_geo to keep inference inside the US only, all token prices are multiplied by 1.1. Global routing is billed at the standard price.

How much does Sonnet 5.5 prompt caching cost?

Prompt caching stores the start of a prompt, such as a long system prompt or a set of source files, so that later requests can reuse it. You pay once to write the cache. After that, each request that reads it pays a much lower rate for that part of the prompt.

Sonnet 5.5 offers two cache lifetimes, also called TTL (time to live). A write that lives five minutes costs $2.50 per million tokens. A write that lives one hour costs $4. The claude.com pricing page shows only the first figure, because its caching prices assume the five-minute TTL.

A cache read costs $0.10 per million tokens. That is 5% of the base input price. The general rule on Anthropic's pricing page is that a cache hit costs 10% of input, but the page lists exceptions: Sonnet 5.5 and Opus 5.5 are priced at 5%, and Fable 5.1 and Mythos 5.1 at 2.5%. So do not apply the 10% rule here. Sonnet 5.5 has the same prices as Sonnet 5 in every other column, and its cache read is half the Sonnet 5 rate.

There is a floor. The minimum cacheable prompt length on Sonnet 5.5 is 512 tokens, so caching only helps when the repeated part is at least 512 tokens long. A shorter prompt is processed without caching, and no error is returned.

Here is what caching does to a 40,000-token prefix that is sent with every request, for example a system prompt plus the files of a small project. These figures are arithmetic from the rates above.

[
  {
    "label": "2 calls, no cache",
    "input_cost_usd": "0.16"
  },
  {
    "label": "2 calls, 5-minute cache",
    "input_cost_usd": "0.104"
  },
  {
    "label": "10 calls, no cache",
    "input_cost_usd": "0.80"
  },
  {
    "label": "10 calls, 5-minute cache",
    "input_cost_usd": "0.136"
  }
]

Two calls with no cache cost $0.16 for that prefix. With a five-minute cache they cost $0.104, because the first call pays the write rate and the second pays the read rate. So the cache pays for itself on the first reuse. The gap grows with every call after that. Ten calls cost $0.80 without a cache and $0.136 with one, as long as the cache entry has not expired between calls.

The one-hour write costs twice the base input rate, so it needs more reuse before it pays back. Use it when the gap between calls is longer than five minutes. A common case is a person who reads each answer before asking the next question.

What does a coding session cost on Sonnet 5.5?

Take a session that uses 50,000 input tokens and 15,000 output tokens in total. Anthropic uses the same token counts for its own Opus 5.5 example on the pricing page. The input is everything sent to the model across the session, including context that is sent again on each turn.

[
  {
    "label": "Sonnet 5.5, standard",
    "input_usd": "0.10",
    "output_usd": "0.15",
    "total_usd": "0.25"
  },
  {
    "label": "Sonnet 5.5, Batch",
    "input_usd": "0.05",
    "output_usd": "0.075",
    "total_usd": "0.125"
  },
  {
    "label": "Opus 5.5, standard",
    "input_usd": "0.20",
    "output_usd": "0.30",
    "total_usd": "0.50"
  },
  {
    "label": "Haiku 5.5, standard",
    "input_usd": "0.005",
    "output_usd": "0.0075",
    "total_usd": "0.0125"
  }
]

At standard rates the session costs $0.25 on Sonnet 5.5: $0.10 for input and $0.15 for output. The same tokens cost $0.50 on Opus 5.5. That is exactly twice as much, because both Opus rates are double the Sonnet rates. Sent through the Batch API, the Sonnet session costs $0.125. On Haiku 5.5 it costs $0.0125, because no single prompt in it can pass Haiku's 100,000-token price step.

Anthropic's own version of this example adds a session runtime line of $0.08 per hour. That line is there because the example sits in the section of the page about a hosted agent product with a per-hour fee. The figures above are token charges only.

Batch requests are processed asynchronously, so they fit jobs where nobody waits on the answer, such as a nightly review of the day's commits. They do not fit an interactive coding session. The Batch discount also stacks with the cache prices.

Count your own tokens before you trust any estimate, this one included. Sonnet 5.5 runs at effort high by default, and Anthropic's help centre says higher effort means more thorough answers that use more tokens. How Claude Code spends tokens shows where they go in a real session, and the difference between Claude effort levels explains the setting that changes the count.

Can you use Sonnet 5.5 on Free, Pro, Max, Ultra or Team?

The claude.com pricing page lists Sonnet on all five plans. It does not print a version number in the plan lists. The help centre's context-window article does name Sonnet 5.5 for the paid plans (Pro, Max, Ultra, Team and Enterprise), but we found no help centre article that names it for the Free plan. On Free, the reliable check is the model picker in the Claude app, signed in to your own account.

What the plan pages say, as of October 31, 2026:

• Free ($0) lists Sonnet and Haiku.
• Pro costs $20 a month billed monthly, or $17 a month billed annually ($200 up front). It lists Sonnet and Opus.
• Max costs $100 a month, with five times Pro's usage every five hours and every week, plus $100 a month in API credits. It lists Sonnet and Opus.
• Ultra costs $200 a month, with 10 times Pro's usage every week, plus $200 a month in API credits. It lists Sonnet and Opus. Ultra replaces the former Max 20x plan, and existing subscribers move automatically with no price change. Max and Ultra are billed monthly only; these are web prices, and mobile prices may vary.
• Team Standard seats cost $25 a month billed monthly or $20 billed annually. Premium seats cost $125 or $100. The Team help article promises "access to all available models", with a minimum of 2 members and a maximum of 150 seats.

The help centre does confirm two Sonnet 5.5 details inside the app. The effort selector is available for Sonnet 5.5. Thinking cannot be turned off for it. On paid plans its context window is 1M tokens in chat and in Claude Code, and Cowork gets the same 1M.

For what each paid plan includes beyond the model list, see Claude Pro price and limits and how to choose between Claude Max and Ultra.

How does Sonnet 5.5 use your plan limits?

Pro and Max have a session limit that resets every five hours, plus a weekly limit. The Max weekly limit applies across all models. Ultra has a weekly limit across all models, with no five-hour limit for verified accounts; accounts not yet verified have a temporary limit. Max and Ultra also have a separate weekly Fable limit. Usage in Claude and Claude Code counts toward the same limits, so a long Claude Code run on Sonnet 5.5 also shrinks what is left for chat.

How fast a message uses the limit depends on the model you choose and the effort level. We found no per-model number for Sonnet 5.5 in the help centre. Anthropic's Academy guide gives only an order: Haiku is the lightest, Sonnet is moderate, Opus is heavy, and Fable uses the most. Moving a task from Opus to Sonnet therefore stretches your limit, but no published figure says by how much. If you run out anyway, what to do when you reach a Claude limit covers the options.

One app behaviour can change the model under you. Automatic model switching is on by default for Sonnet 5.5. If Anthropic's cyber classifiers flag a request, Claude runs it again on Sonnet 5, and the model picker stays on Sonnet 5 for the rest of that chat. You can turn the switching off. On the API nothing switches unless you opt in and configure fallbacks. A fallback response is billed at Sonnet 5 rates, with a credit for the cache miss.

Should you use Sonnet 5.5 on the API or on a plan?

It is the same model either way, billed in two different ways. The API charges for each token, so a quiet month costs little and a busy month has no ceiling. A plan charges a flat monthly fee, and weekly limits plus any five-hour limit cap the usage that fee includes. If you turn on usage credits on a paid plan, requests past those limits switch to pay-as-you-go at standard API rates and keep working, up to a monthly spend limit you can set. The monthly API credits included with Max and Ultra are separate and do not cover usage credits in Claude, Claude Code or Cowork. Use the session chart above to price your own month of API use, then compare it with a plan in Claude API versus subscription cost.

When is Sonnet 5.5 the right pick over Opus 5.5 and Haiku 5.5?

Anthropic's model guide calls Sonnet 5.5 "the best combination of speed and intelligence". It sits between Opus 5.5, which has moderate latency, and Haiku 5.5, which is the fastest. The guide lists Sonnet 5.5 for code generation, data analysis, content creation, visual understanding and agentic tool use.

[
  {
    "label": "Opus 5.5",
    "input_usd_per_mtok": 4,
    "output_usd_per_mtok": 20
  },
  {
    "label": "Sonnet 5.5",
    "input_usd_per_mtok": 2,
    "output_usd_per_mtok": 10
  },
  {
    "label": "Haiku 5.5, prompt up to 100K",
    "input_usd_per_mtok": "0.10",
    "output_usd_per_mtok": "0.50"
  },
  {
    "label": "Haiku 5.5, prompt over 100K",
    "input_usd_per_mtok": "0.50",
    "output_usd_per_mtok": "2.50"
  }
]

Pick Sonnet 5.5 when the work is everyday coding or agent steps, and the volume is high enough that the bill matters. At $2 in and $10 out, it costs half as much per token as Opus 5.5 at $4 and $20.

Pick Opus 5.5 for the jobs the guide names for it, such as multi-hour autonomous coding agents and large-scale refactoring. On a long run with nobody watching, one wrong turn can cost more time than the token savings are worth. Note that the same guide says most workloads start with Opus 5.5. Starting with Sonnet is a decision you make for cost, and Anthropic's own default is Opus.

Pick Haiku 5.5 for real-time replies and for subagent tasks that run many times. Up to 100,000 prompt tokens it costs $0.10 in and $0.50 out. Above that, its price rises to $0.50 and $2.50, so very long prompts close part of the gap with Sonnet.

The full three-way comparison is in which Claude model to pick: Opus, Sonnet or Haiku. The matching price pages for the other two models are Claude Opus 5.5 price and plans and Claude Haiku 5.5 price and uses.

What changes in your API code on Sonnet 5.5?

Check your request code for these settings before you change the model ID to claude-sonnet-5-5. Each one returns an error on Sonnet 5.5:

• thinking: {"type": "disabled"} returns a 400 invalid_request_error. The error message points you to between_tools, which is the setting to use instead.
• Forced tool use, meaning tool_choice set to any or to one named tool, returns an error.
• Setting temperature, top_p or top_k returns a 400 error.
• Remove these settings in the same change that switches the model ID, or the first request after the switch fails.

The default effort is the other change that touches cost. Sonnet 5.5 defaults to high, while Opus 5.5 and Haiku 5.5 default to medium. So a prompt moved from either of them can produce more output tokens on Sonnet 5.5 unless you set the effort yourself. On the Batch API, output can reach 300K tokens per request with a beta header. The normal limit is 128K.

FAQ

How much does Claude Sonnet 5.5 cost per million tokens?

On the Claude API, Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, checked on October 27, 2026. The Batch API halves both, to $1 and $5. A five-minute cache write costs $2.50, a one-hour cache write $4, and a cache read $0.10 per million tokens. The price is the same at any prompt length inside the 1M-token context window.

Is Claude Sonnet 5.5 cheaper than Opus 5.5?

Yes. Per token it costs half as much: Opus 5.5 is $4 in and $20 out per million tokens. A session of 50,000 input and 15,000 output tokens costs $0.25 on Sonnet 5.5 and $0.50 on Opus 5.5. On a Claude plan, Anthropic's Academy guide says Sonnet uses less of your limit than Opus, but it gives no exact ratio.

Can I use Claude Sonnet 5.5 on the free plan?

The claude.com pricing page lists Sonnet and Haiku on the Free plan, but it does not name a Sonnet version. To see whether Sonnet 5.5 is offered on your account, open the model picker in the Claude app. Pro, Max, Ultra and Team also list Sonnet, and the help centre gives Sonnet 5.5 a 1M-token context window on paid plans.

Did the Sonnet price go up to $3 and $15 on 1 September 2026?

No. Anthropic cancelled the scheduled September 1, 2026 price increase for Sonnet 5 and made its introductory price the standard price, so Sonnet 5 stays at $2 input and $10 output per million tokens. Sonnet 5.5 is priced at $2 and $10 too, and its row carries no note about a rise. The older Sonnet 4.6 and Sonnet 4.5 still cost $3 and $15.