OpenAI-compatible and Anthropic-compatible APIs, explained

The official references are the Chat Completions, Responses and Messages docs. A compatible server accepts those request bodies and returns the same fields: choices and usage for Chat Completions, output for Responses, content blocks for Messages.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.openscale.so/v1",
    api_key="YOUR_OPENSCALE_API_KEY",
)
reply = client.chat.completions.create(
    model="deepseek/deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Hello"}],
)
print(reply.choices[0].message.content)

For an Anthropic SDK client, the base URL is https://api.openscale.so/anthropic, and the SDK appends /v1/messages itself. The key and the model name change the same way.

• Covered: messages, streaming, tool calls, JSON output, usage counts and the error shape of the format.
• Model-specific: each model supports its own set of settings. A good compatible API refuses a value the model cannot take with a clear 400 that names the parameter, instead of ignoring it.
• Provider-specific: hosted tools tied to one company's platform (built-in web search, for example) are not part of the format.

Before you switch, read the target model's limits. On OpenScale, GET /v1/models lists each model's context length, maximum output and supported parameters. DeepSeek V4.1 Flash accepts up to 384,000 output tokens per request and GLM 5.3 Flash up to 131,072.

Coding agents pick one format. Some speak Anthropic Messages, others Chat Completions or Responses. Point the agent at the matching base URL and it runs on an open model with no other change. The setup guide has a tested config for each agent we support.