Claude Opus 5.5 API: A Working curl and Python Quickstart

Claude Opus 5.5 is Anthropic's newest agentic-coding and long-context reasoning model, and hiapi exposes it the same day it lands on the platform. Unlike hiapi's image, video, and audio models — which all go through the async POST /v1/tasks → poll pattern — Claude Opus 5.5 is a text-only chat model, so you call it with a single, synchronous request to /v1/chat/completions. This guide gives you a working curl call, the same call in Python, and the production details (reasoning effort, streaming, error handling) you need before you ship it.

What you need

• A hiapi account and an API key. Create one from API Keys — every hiapi model, including Claude Opus 5.5, shares the same key.
• No SDK install required. Claude Opus 5.5 uses hiapi's OpenAI-compatible Chat Completions format, so any HTTP client (curl, requests, fetch) works out of the box.

The minimal working request (curl)

curl -X POST "https://api.hiapi.ai/v1/chat/completions" \
  -H "Authorization: Bearer sk-YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-5-5",
    "messages": [
      { "role": "user", "content": "Give me three tips for reviewing a Python pull request." }
    ],
    "stream": false
  }'

The model id is passed bare — no route suffix, no provider prefix. A successful call returns a standard chat-completion object:

{
  "id": "msg_...",
  "object": "chat.completion",
  "model": "claude-opus-5-5",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "1. ..." },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 13,
    "completion_tokens": 42,
    "total_tokens": 55,
    "usage_semantic": "openai",
    "usage_source": "anthropic"
  }
}

Read the answer from choices[0].message.content — there's no output[0].url to poll for, and no task id to track.

The same flow in Python

import requests

response = requests.post(
    "https://api.hiapi.ai/v1/chat/completions",
    headers={
        "Authorization": "Bearer sk-YOUR_API_KEY",
        "Content-Type": "application/json",
    },
    json={
        "model": "claude-opus-5-5",
        "messages": [
            {"role": "user", "content": "Give me three tips for reviewing a Python pull request."}
        ],
        "stream": False,
    },
)
response.raise_for_status()
data = response.json()
print(data["choices"][0]["message"]["content"])

Because the endpoint speaks the OpenAI Chat Completions wire format, this same payload shape also works unmodified through any HTTP client in any language — there's nothing hiapi-specific to learn beyond the base URL and the bearer key.

Reasoning effort, thinking, and streaming

Claude Opus 5.5 keeps extended thinking always on — you can't disable it or hand it a manual token budget. What you control is how much of it the model does, via output_config.effort:

{
  "model": "claude-opus-5-5",
  "messages": [{ "role": "user", "content": "..." }],
  "output_config": { "effort": "medium" }
}

effort accepts low, medium (the default if you omit the field), high, xhigh, or max — going from fast/cheap answers to deep, deliberate ones. The response's message includes a reasoning_details array holding the model's internal thinking trace; for a simple one-shot call you can safely ignore it and just read message.content.

For long answers, set "stream": true to get incremental output instead of waiting for the full response:

curl -N https://api.hiapi.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-opus-5-5","messages":[{"role":"user","content":"..."}],"stream":true}'

Each chunk is a data: {...} line shaped like chat.completion.chunk, with the incremental text in choices[0].delta.content. Keep reading chunks until one arrives with a non-null finish_reason (normally "stop") — that's the end of the response.

Claude Opus 5.5 also supports tool/function calling and has a 1M-token context window with up to 128K tokens of output, so it's built for large-codebase or long-document tasks, not just short Q&A.

Production notes

• Auth errors are 401, not 400. A missing or invalid key returns:

  { "error": { "code": "permission_denied", "message": "This API key is invalid...", "type": "hiapi_error", "request_id": "..." } }
  

  Check for HTTP 401 specifically and surface the request_id if you need to contact support — see the full invalid API key troubleshooting guide for other causes of this error.

• There's no callback/webhook option here. That pattern exists for hiapi's async task-based models (image, video, audio); Chat Completions is request/response only, so stream: true is your only alternative to a single blocking call.

• This call is idempotent by nature — there's no task id to accidentally double-create, so retries on timeout are safe as long as you're comfortable re-running the same prompt.

• Watch your rate limits on high-concurrency workloads; see how to handle a 429 rate-limit error for backoff strategy.

• Token pricing (input, output, and cached tokens are billed separately) is kept current on the Claude Opus 5.5 model page and the pricing page — don't hardcode numbers in your billing logic, read them from there.

Related reading

• Claude Opus 5.5 model page — full spec sheet and live pricing
• Claude Sonnet 4.6 API guide — a lighter, faster Claude option on the same Chat Completions format
• Building an OpenAI-compatible proxy in Python — useful if you're fronting multiple chat models behind one internal API
• Manage your API keys

FAQ

Does the Claude Opus 5.5 API support streaming? Yes — set "stream": true and read the data: {...} chunks as described above.

Do I need an API key to call Claude Opus 5.5 in production? Yes. Every request must carry Authorization: Bearer sk-<your key>; there's no keyless or anonymous access to the production API.

Can I turn off Claude Opus 5.5's thinking to save tokens? No. Thinking is always on for this model; the only control you have is the output_config.effort level (low through max), which trades speed/cost for depth.

Is Claude Opus 5.5 called through /v1/tasks like hiapi's image models? No. Only hiapi's image, video, and audio models use the async task-and-poll pattern. Claude Opus 5.5 is a text model on /v1/chat/completions and returns its answer in the same response.

What's the context window and max output length? 1,000,000 tokens of context and up to 128,000 tokens of output — large enough for whole-repo or long-document prompts.

Does Claude Opus 5.5 support tool/function calling? Yes, alongside streaming and the OpenAI-compatible message format.