Models
The catalogue below is live. It is the same list the API serves, so a model that appears here is one you can call right now, and one that does not will be refused by name.
Call GET /v1/models with your key for the same list in
machine-readable form — that is the better source if you generate client
code or validate model names at build time.
| Model | Context tokens | Max output tokens |
|---|---|---|
|
Haiku 4.5
Haiku 4.5
|
200,000 | 64,000 |
|
Opus 5
Opus 5
|
1,000,000 | 128,000 |
|
Sonnet 5
Sonnet 5
|
1,000,000 | 128,000 |
The ceiling on everything a single request can carry at once: your prompt, the conversation so far, and the response together. Exceed it and the request is refused rather than quietly truncated, because a silently shortened prompt is a wrong answer you would have no way to notice.
The most a model will generate in one response, separate from the context window and usually far smaller than it. It is the ceiling you set on a response, so the length of what comes back is something you decide rather than something you discover.
Larger models for reasoning and long documents, smaller ones for classification, extraction and anything high-volume. If you are unsure, describe the workload and we will tell you where the trade-off actually falls.
Any OpenAI-compatible client works. The model id is the alias — the lowercase string under each name above.
from openai import OpenAI
client = OpenAI(
base_url="https://api.claudcli.com/v1",
api_key="sk-your-key-here",
)
response = client.chat.completions.create(
model="Haiku 4.5",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)Parameter reference, streaming and error handling are in the documentation. What limits apply, if any, are listed there too — by default there are none, and a request stops only when the token balance does.
Describe what you are building and we will point you at the model that does it for the least money.