Claude Opus 5.5 overview: API access, pricing and testing on your tasks

Claude Opus 5.5 is the top Anthropic model in the Zukko catalog: 1M-token context, callable in Anthropic or OpenAI format. Below: how to connect Claude Opus 5.5 via the API, what a request costs (with formulas), how to test the model on your own tasks, and how to keep spending under control.

How to call Claude Opus 5.5

Model ID: claude-opus-5-5. It is available in both formats; the gateway converts requests for you.

Anthropic SDK (address without /v1; the SDK adds it):

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.zukko.pro",
    api_key="fetch_your_key",
)

response = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Describe the architecture of a payment processing service"}
    ],
)
print("".join(b.text for b in response.content if b.type == "text"))

Collect text from text blocks: the response may contain other block types (such as thinking), so response.content[0].text is not a reliable way to read it.

OpenAI-compatible format (address with /v1):

from openai import OpenAI

client = OpenAI(
    base_url="https://api.zukko.pro/v1",
    api_key="fetch_your_key",
)

response = client.chat.completions.create(
    model="claude-opus-5-5",
    messages=[
        {"role": "user", "content": "Describe the architecture of a payment processing service"}
    ],
)
print(response.choices[0].message.content)

Claude Code (environment variables):

export ANTHROPIC_BASE_URL=https://api.zukko.pro
export ANTHROPIC_AUTH_TOKEN=fetch_your_key
export ANTHROPIC_MODEL=claude-opus-5-5

More on choosing a format: Anthropic or OpenAI format↗.

One technical detail: according to the Anthropic Messages API reference↗, models released after Claude Opus 4.6 do not accept non-default temperature and top_p values; such requests get a 400 error. If you are porting old code, drop these parameters and steer the output with prompt instructions.

Claude Opus 5.5 pricing vs the official API

Prices per 1M tokens:

Example: 10 documents of 50K tokens each (500K input tokens) and a 10K-token answer.

A note on caching. At the vendor, prompt cache reads cost about 10% of the input price, while Zukko bills cached tokens at the regular input price. If an agent sends the same large context over and over, compare against the cached price. For example, a repeated read of 800K context tokens:

There is still a difference, but it is much smaller than a straight price-list comparison suggests.

Also note: on Zukko, Claude Opus 5.5 and Claude Sonnet 5 cost the same ($0.30 / $0.30). So choosing between them is not about per-token price but about which model solves your task and how many tokens it spends on an answer.

How to tell whether you need Opus 5.5

You will not find "this model is smarter" rankings here: we have no measurements of our own, and third-party benchmarks rarely match your task. It is more reliable to test it yourself, and it takes about an hour.

1. Collect 20–30 real requests from your work, including hard and edge cases.
2. For each, write down a success criterion: what must be in the answer, what must not, which format.
3. Run the same requests through claude-opus-5-5 and claude-sonnet-5 (you can add glm-5.3-flash as the cheapest baseline).
4. Grade blind: save answers without the model name, shuffle them, mark pass / fail.
5. Compute the pass rate and the average tokens per answer for each model.

A script for steps 3 and 5:

import json, random
import anthropic

client = anthropic.Anthropic(base_url="https://api.zukko.pro", api_key="fetch_your_key")
models = ["claude-opus-5-5", "claude-sonnet-5"]
prompts = [line.strip() for line in open("prompts.txt", encoding="utf-8") if line.strip()]

rows = []
for prompt in prompts:
    for model in models:
        r = client.messages.create(
            model=model,
            max_tokens=4096,
            messages=[{"role": "user", "content": prompt}],

For blind grading, open answers.jsonl with the model column hidden (for example, in a spreadsheet) and reveal it only after grading. If both models pass equally often, pick the one that spends fewer tokens. If Opus 5.5 passes your criteria noticeably more often, use it.

Spending control: limits and keys

In the dashboard↗ you can create a separate key for a specific service and set:

• a daily limit in $: once reached, requests with this key are rejected until the next day;
• a monthly limit in $: the same for the month;
• allowed models (allowed_models): for example, only claude-opus-5-5 and claude-sonnet-5.

Error codes worth handling:

import anthropic

try:
    response = client.messages.create(
        model="claude-opus-5-5",
        max_tokens=4096,
        messages=[{"role": "user", "content": "..."}],
    )
except anthropic.APIStatusError as e:
    if e.status_code == 402:
        print("Insufficient balance: top up your account")
    elif e.status_code == 403:
        print("Model not allowed for this key (model_not_allowed_for_key)")
    elif e.status_code == 429:
        print("Too many requests: retry with a pause")
    else:
        raise

More on key limits in the docs↗. The model set changes: the current list is at GET https://api.zukko.pro/v1/models or on the "Models" page in the dashboard.

When a 1M-token context helps

A large window is useful when material cannot be split into parts without losing something:

• analyzing a service's whole codebase;
• long correspondence or minutes where links between early and late messages matter;
• a set of documents that must be cross-checked against each other.

Before sending everything at once, do the math: 800K input tokens is 800,000 × $0.30 / 1,000,000 = $0.24 per request. If you will ask many questions about the same context, multiply by their number and keep the caching note above in mind. You can find out how many tokens your text takes in advance with count_tokens↗.

FAQ

Is Opus 5.5 slower than Sonnet 5?

We do not publish speed measurements. Measure it yourself: send identical streaming requests and record time to first token and total response time for each model. Compare the median over 10–20 requests, not single runs.

Can I switch between Opus and Sonnet without code changes?

Yes. Just replace model="claude-opus-5-5" with model="claude-sonnet-5". It is handier to keep the model name in an environment variable.

How do I know I don't need Opus?

Run your task set using the method above. If a cheaper model (glm-5.3-flash or gpt-5.6-luna) passes your criteria just as often, pay less. If not, compare Opus 5.5 and Sonnet 5: on Zukko they cost the same, so the pass rate decides.

What is the maximum request size?

The model context is up to 1M tokens. The gateway rejects an oversized request with a 413. Response length is set by max_tokens; for long answers use streaming to avoid timeouts.