Anthropic SDK / Messages format

Besides the OpenAI format, AI Nexus offers the Anthropic-native POST /v1/messages endpoint. You can use the official Anthropic SDK directly, and it is also how Claude Code connects.

Differences from the OpenAI formatCopy Anchor Link

ExampleCopy Anchor Link

Replace your-model-id in the samples with a model ID from the model square (or call GET /v1/models).

import os

import anthropic

client = anthropic.Anthropic(
    api_key=os.environ["MYAIEXP_API_KEY"],
    base_url="https://ai.myaiexp.com",  # note: the Anthropic SDK base_url has no /v1
)

message = client.messages.create(
    model="your-model-id",  # replace with a model ID from the model square
    max_tokens=256,
    messages=[{"role": "user", "content": "Say hello in one sentence."}],
)

print(message.content[0].text)

Install the SDK:

pip install anthropic

If your tool only supports Bearer authentication, replace x-api-key: $MYAIEXP_API_KEY in the curl with Authorization: Bearer $MYAIEXP_API_KEY; in the Anthropic SDK this is the auth_token / authToken option.

Extended thinkingCopy Anchor Link

Models that support extended thinking can think before they answer: send the thinking parameter in the request. thinking is a standard parameter of the Anthropic Messages API and is passed through as defined in Anthropic's API. For its format, refer to Anthropic's documentation; which models support extended thinking is listed on the model square.

Notes:

• Thinking is returned as type: "thinking" content blocks before the answer. Don't read content[0] blindly; filter for type == "text".
• Thinking tokens count as output tokens and are billed as output. Leave enough room in max_tokens, or the response may be cut off mid-thought.
• Thinking takes time, so pair it with streaming.

Improve cache hit rateCopy Anchor Link

For multi-turn conversations or long system prompts, add cache_control to content blocks to enable prompt caching; repeated input is billed at the vendor's cache-read list price. See Prompt caching.

Choose the right modelCopy Anchor Link

• /v1/messages only accepts IDs of models that support the Claude-native format; other models return an error. Pick a model on the model square that supports /v1/messages and use its model ID.
• Model IDs are case-sensitive. The models your key can use are those returned by GET /v1/models; the full list and prices are on the model square.
• If you set a model restriction when creating the key, make sure the model you need is in the allowed list. See API Keys.

Supported featuresCopy Anchor Link