APIMart
Back to all models

Claude Opus 5.5 API

AnthropicModel ID claude-opus-5-5Text modelReleased September 22, 2026

Use Opus 5.5 to roll a dependency upgrade across a whole codebase, find the endpoint that skips an ownership check, or reconcile numbers that don’t match across reports.

Max input
1,000,000 tokens
Max output
128,000 tokens
Input → output
Text, Image → Text
Knowledge cutoff
June 2026
  • Image input
  • Streaming
  • Cached input pricing
  • OpenAI SDK compatible

Claude Opus 5.5 API use cases

Each use case comes with an illustrative example: source material, a prompt and a hand-written sample answer, not actual model output. Copy the prompt, swap in your own material and try it on Claude Opus 5.5.

Try Claude Opus 5.5 with your own material

Codebase-wide migrations

Send the upgrade notes with a few typical call sites, and have Claude Opus 5.5 find every pattern that needs to change, show the new form, and plan the order of batched commits. For big changes, run the related tests on each batch before merging.

Example: Which calls does this dependency upgrade break?

Material

package.json
"date-lib": "^2.4.0" upgraded to "^3.0.0".
Upgrade notes
In v3, the second argument of format() changes from a string to an object { pattern }.
src/report.ts
const month = format(date, "YYYY-MM");

PromptUsing the upgrade notes, find the calls in the sample code that need to change, write the updated code, and lay out how to finish the migration across the repo in batches.

Sample answer · written by hand to show the reasoning

The call in report.ts becomes format(date, { pattern: "YYYY-MM" }). Search the repo for every format( call and batch them by directory. After each batch, run the report tests and confirm the date strings match the pre-upgrade output before merging the next one.

About the Claude Opus 5.5 API

  • First model in the Claude 5.5 family

    Anthropic released Claude Opus 5.5 on September 22, 2026 for long-running agentic coding and knowledge work. Its docs recommend starting with it when you’re unsure which model to use.

  • Built for codebase-scale long tasks

    Anthropic says it’s especially strong at long, sprawling work such as codebase-wide migrations and audits. Move everyday high-volume tasks to Claude Sonnet 5.5, at half the official price of Claude Opus 5.5.

  • Thinking always on, depth set by effort

    Claude Opus 5.5 can’t turn thinking off; the default effort in the official API is medium. In Anthropic’s tests at default settings, typical tasks cost 40% less to run than on Opus 5 and output is over 30% faster.

  • Claude Opus 5.5 API key and access

    You don’t need a separate Claude Opus 5.5 API key: one APIMart key works for every model. Requests go through an OpenAI Chat Completions-compatible endpoint with the model ID claude-opus-5-5.

Claude Opus 5.5 official benchmark results

Selected results Anthropic published at launch.

  • Terminal-Bench 4.0 (Xhigh)

    66.4%

    Real development and ops tasks in a terminal. 14.1 points above Opus 5 (52.3%) and ahead of GPT-6 Astra (57.9%).

  • CursorBench 4.0

    57.8%

    Coding tasks in Cursor, scored at max effort; 52.5% at the default medium effort. Opus 5 scores 46.6%.

  • FrontierCode v1.1

    54.4%

    Whether changes to real codebases meet maintainers’ bar for merging. Ahead of GPT-6 Astra (53.3%) and Opus 5 (48.0%).

  • GDPval-AA v2.1

    1,846 Elo

    Knowledge work across industries. Ahead of Opus 5 (1,708). Anthropic says at default effort it beats GPT-6 Astra at max effort for about a fifth of the cost per task. Elo is a relative rating from head-to-head comparisons; higher is better.

  • Humanity's Last Exam (with tools)

    67.7%

    Hard questions across many disciplines, with tools allowed. Ahead of Claude Fable 5.1 (65.6%) and Opus 5 (63.6%).

  • OSWorld 2.1 (partial)

    81.8%

    Completing tasks by operating a computer in a real desktop environment. 7.8 points above Opus 5 (74.0%).

How to call the Claude Opus 5.5 API

To call Claude Opus 5.5, you only need an APIMart API key and your existing OpenAI SDK, or send plain HTTP requests.

POST /v1/chat/completions
curl https://api.apimart.ai/v1/chat/completions \
  -H "Authorization: Bearer $APIMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "claude-opus-5-5",
  "messages": [
    {
      "role": "user",
      "content": "Describe how to design a clear API in three sentences."
    }
  ],
  "stream": false
}'

Claude Opus 5.5 access details

Model IDclaude-opus-5-5
Base URLhttps://api.apimart.ai/v1
Endpoint/chat/completions

Common parameters

model
Required
messages
Required
stream
Optional
max_tokens
Optional
temperature
Optional
All request parameters, response fields and error codes

Claude Opus 5.5 API request parameters

Claude Opus 5.5 uses the general chat endpoint, POST /v1/chat/completions, with a JSON body. model and messages are required; every other parameter is optional.

ParameterTypeRequiredDescription
modelstringYesModel ID. Use claude-opus-5-5 for this model.
messagesarrayYesConversation messages. Each message has a role (system, user, or assistant) and content. Include earlier messages for multi-turn conversations.
streambooleanNoWhether to stream the output. true returns chunks over SSE, false returns the full response at once. Default: true.
temperaturenumberNoOutput randomness, 0–2. Default: 1.0. Lower values give more deterministic output.
top_pnumberNoNucleus sampling, 0–1. Default: 1.0. Adjust this or temperature, not both.
max_tokensintegerNoMaximum number of tokens to generate for the reply.
frequency_penaltynumberNoFrequency penalty, -2.0 to 2.0. Default: 0. Positive values reduce repeated wording.
presence_penaltynumberNoPresence penalty, -2.0 to 2.0. Default: 0. Positive values encourage new topics.
stopstring | arrayNoUp to 4 stop sequences. Generation stops when any of them appears.
nintegerNoNumber of replies to generate. Default: 1.

Response fields

The response contains the fields below. Token usage is returned in usage, which you can use to check costs.

FieldTypeDescription
idstringUnique identifier for the request.
modelstringThe model that handled the request.
choicesarrayList of replies. Each item contains message (role and content) and finish_reason.
usageobjectToken usage: prompt_tokens, completion_tokens, and total_tokens.

Error codes

Failed requests return an error object with code, message, and type.

StatusTypeMeaning
400invalid_request_errorInvalid request parameters.
401authentication_errorThe API key is invalid or missing.
402payment_requiredInsufficient account balance.
403permission_errorNo permission to access this resource.
429rate_limit_errorToo many requests. Retry later.
500server_errorInternal server error.
502bad_gatewayGateway error. The service is temporarily unavailable.

Claude Opus 5.5 API pricing

The Claude Opus 5.5 API is billed by token usage, and prices depend on your membership tier. Prices below are per million tokens; 1 US dollar = 10 credits.

ItemGoldCurrent pricePlatinumDiamondOfficial priceYou save
Input32Credits~$3.230.4Credits~$3.0428.8Credits~$2.8840Credits~$420%
Cached input1.6Credits~$0.161.52Credits~$0.1521.44Credits~$0.1442Credits~$0.220%
Output160Credits~$16152Credits~$15.2144Credits~$14.4200Credits~$2020%

Prices are for reference only. View full pricing

Why call Claude Opus 5.5 API through APIMart

  • 20% below official pricing

    Claude Opus 5.5 API pricing is billed by token usage. Gold to Diamond member prices are 20%–28% below Anthropic's official prices, and cached input costs just $0.16 per 1M tokens.

    See membership prices
  • One API key for 500+ models

    Claude Opus 5.5, the rest of its family and 500+ models from GPT, Claude, Gemini and more share one API key and one account balance. Switching models is just a change to the model parameter.

  • Change two lines of existing code

    If you already use the OpenAI SDK, point base_url at https://api.apimart.ai/v1 and swap in your APIMart API key. Requests and responses keep the same format.

  • Try it online before you code

    Try Claude Opus 5.5 with your own material in the Playground, no code or setup needed, and integrate once you like the results.

    Open Playground

Which Anthropic model should you choose?

Base prices per 1M tokens (USD). Your Claude Opus 5.5 API key works for every model in the table; just change the model parameter.

Differences between Anthropic models
ComparisonClaude Opus 5.5 CurrentClaude Sonnet 5.5
Best forAgentic coding, Large-scale refactoring, Knowledge workEveryday coding, Agentic tasks, Data analysis
Input price$3.2$1.6
Output price$16$8

Claude Opus 5.5 FAQ

How is Claude Opus 5.5 different from Claude Sonnet 5.5?

Both belong to Anthropic’s Claude 5.5 family and have the same context window and max output. Opus 5.5 targets long-running agentic coding and knowledge work, with adaptive thinking always on; Sonnet 5.5 is faster and costs half as much at official prices, which suits everyday coding and high-volume tasks. On APIMart both share one endpoint and API key, so you only change the model parameter.

How is Claude Opus 5.5 different from Opus 5?

Claude Opus 5.5 is the next generation after Opus 5. According to Anthropic, Claude Opus 5.5 pricing is 20% lower than Opus 5 on input and output, the cache read price drops from 10% to 5% of the input price, and the context window (1,000,000 tokens) and max output (128,000 tokens) stay the same. In Anthropic’s testing at default settings, typical tasks cost 40% less to run and output is over 30% faster. To switch from Opus 5, set model to claude-opus-5-5 and rerun your own prompts to check the output.

When was Claude Opus 5.5 released?

Anthropic released Claude Opus 5.5 on September 22, 2026, with a reliable knowledge cutoff of June 2026. It is available on APIMart under the model ID claude-opus-5-5.

Can I turn off thinking on Claude Opus 5.5?

No. According to Anthropic’s documentation, Claude Opus 5.5 uses adaptive thinking that is always on, and the model decides how much to think for each task. In Anthropic’s official API the effort parameter controls depth, with medium as the default. For faster, cheaper replies, switch to Claude Sonnet 5.5.

How can I try this model online?

Click Start chatting or Open Playground to visit the existing APIMart Playground. Select Claude Opus 5.5 in the text category to start exploring.

Which model ID does the Claude Opus 5.5 API use?

Set model to claude-opus-5-5. Copy the exact ID from the page heading or the integration details panel.

Is it compatible with the OpenAI SDK?

You can keep the OpenAI SDK. Set base_url to https://api.apimart.ai/v1, provide your APIMart API key, and use the model ID on this page.

What is the context window of Claude Opus 5.5?

On APIMart, Claude Opus 5.5 accepts up to 1,000,000 input tokens per request. Your prompt, earlier messages, and file content all count toward it.

How many tokens can it output at once?

Claude Opus 5.5 can output up to 128,000 tokens in a single reply. Use the max_tokens parameter to set a lower limit.

How much does the Claude Opus 5.5 API cost?

The Claude Opus 5.5 API on APIMart is billed by token, and prices depend on your membership tier. Gold: $3.2 per million input tokens and $16 per million output tokens. Diamond: $2.88 per million input tokens and $14.4 per million output tokens. Prices may change; the pricing page has the current rates.

Does it support streaming?

Yes. The general chat endpoint streams over SSE by default. Set stream to false to receive the complete response at once.

How do I get an API key?

Sign in to APIMart and create a key on the API Keys page, then send it in the Authorization: Bearer header. The same key also works for other models on the platform.