Introduction
Claude Sonnet 5.5 is Anthropic's newest Sonnet, released on 28 September 2026, and Anthropic describes it as its best combination of speed and intelligence. It is served here on one OpenAI-compatible endpoint at /v1/chat/completions. A request can carry up to 1,000,000 tokens of text, and one reply can run to 128,000 tokens.
reasoning_effort takes none, low, medium, high, xhigh or max, with medium as the default; none asks the model to skip thinking before the reply, though in long conversations it can still think briefly. The model fixes its own sampling, so temperature and top_p are not accepted. Up to 128 tools can be defined per request, tool_choice takes auto or none, and replies stream over Server-Sent Events.
Key features & highlights
Thinking you set per request
The lever you control is thinking: none asks for no thinking pass before the reply, low and medium keep it short, and high, xhigh and max allow progressively more. A one-line answer and a long plan go through the same integration, and only one field changes between them.
Scenes carried through
The platform adds no filter of its own. In our tests on 1 October 2026 the model refused none of 24 content requests across five levels and wrote every one in full, the Very strong and Maximum levels included.
A million tokens of context
1,000,000 tokens is roughly 555,000 English words: enough for a long story history, a large set of documents or a mid-sized repository sent as written rather than summarised. The model then reads what was actually said, not a retelling that lost the detail you needed.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Ordered conversation turns, oldest first. Roles are system, developer, user, assistant and tool; system and developer messages become the model's instructions. at least 1 items |
max_tokens | No | Caps the reply at 128,000; default 4,096. |
reasoning_effort | No | none, low, medium, high, xhigh or max; medium by default. |
stop | No | Up to 4 sequences, as a string or an array. The reply ends as soon as the model writes one of them, and the matched text itself is left out. at least 1 characters · at least 1 items · at most 4 items |
tools | No | Up to 128 function definitions per request, in the OpenAI function format: a name, a description and JSON-schema parameters for each. at most 128 items |
parallel_tool_calls | No | true by default. Set it to false and the model calls at most one tool per turn, which suits chains where each call depends on the result of the one before. Default: |
tool_choice | No | auto or none. Sonnet 5.5 does not accept a forced tool: to get a specific call, name the tool in the prompt and leave this on auto. |
Pricing
Input $2 / Output $10 per 1 million tokens
Use cases
Interactive fiction and roleplay
Fantasy campaigns, mysteries, starship crews and companion characters all work the same way: persona and world rules in the system message, the story so far in order, and reasoning_effort at none or low for ordinary turns (with max_tokens of at least 4,000, since a long conversation can still make the model think briefly), raised for the scene where several threads have to meet.
The next chapter of a long manuscript
Send the chapters so far and the outline in order, with the voice rules in the system message, and ask for the next chapter. Earlier chapters travel verbatim, so a detail planted in chapter two is still there to pay off in chapter twelve.
Reviews against the whole codebase
Send the repository, the specification and the change in one request and ask for findings you can point to a line for, with reasoning_effort at high when the review has to follow calls across files.
Prompt guide & pro tips
Start at medium and move one step at a time
Anthropic's own default for this model is high, but it recommends starting chat and latency-sensitive work at medium or low. Use none for turns that need no planning (with max_tokens of at least 4,000), and reach for xhigh or max only when a wrong answer costs more than the extra tokens.
Leave room in max_tokens for thinking
Thinking counts toward max_tokens. The default 4,096 suits a conversational reply; for a chapter, a report or a multi-file plan, set it well above the length you expect to read, especially at the higher effort levels.
Ask for a tool call instead of a response format
There is no response_format on this endpoint, and tool_choice cannot force a specific tool. Define a tool whose parameters are the JSON schema you want, leave tool_choice on auto, tell the model to answer by calling it, and validate the arguments that come back. Video and audio parts are not accepted on this endpoint, and a max_tokens above 128,000 is rejected.
Use low, not none, for clean short answers
none asks the model to skip thinking before the reply and is the quickest setting, but in long conversations the model can still think briefly, and that thinking counts against max_tokens; it also sometimes works through the question in the visible text. For a short, clean answer use low; keep none for quick in-character turns and give them at least 4,000 tokens of max_tokens, and set stream to true for long replies so the text appears as it is written.
API quick start
curl --fail-with-body -sN https://api.spicyapi.ai/v1/chat/completions \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $(uuidgen)" \
-d '{
"model": "anthropic/claude-sonnet-5.5/chat",
"messages": [
{
"role": "user",
"content": "Summarise the following release notes into three bullet points."
}
],
"stream": true
}' \
| jq --unbuffered -Rrj 'select(startswith("data: ")) | .[6:] | select(. != "[DONE]") | fromjson | if .error then "\n[error] \(.error.message)\n" else (.choices[0].delta.content? // empty) end'
See the API reference for authentication, the complete schema, polling, and error handling.
