Create a message (Anthropic Messages)

The Anthropic Messages endpoint. Anthropic SDKs and Claude Code use the base URL https://bigmodel.io and append /v1/messages themselves. Anthropic groups serve it natively; OpenAI groups convert it when the group allows Messages dispatch; chat-only groups (DeepSeek, GLM platforms) return 404.

#Request

POST https://bigmodel.io/v1/messages

#Authentication

Send one of these.

#Headers

#Body

Standard Anthropic Messages body.

#Response

#Response fields

With stream: true the response is the Anthropic event stream: message_start, content_block_start, content_block_delta, content_block_stop, message_delta, message_stop.

#Examples

curl https://bigmodel.io/v1/messages \
  -H "Authorization: Bearer $BIGMODEL_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "max_tokens": 256,
    "system": "You are a concise assistant.",
    "messages": [{"role": "user", "content": "Explain HTTP 429 in one sentence."}]
  }'

import os
from anthropic import Anthropic

client = Anthropic(base_url="https://bigmodel.io", api_key=os.environ["BIGMODEL_API_KEY"])

msg = client.messages.create(
    model="glm-5.3",
    max_tokens=256,
    system="You are a concise assistant.",
    messages=[{"role": "user", "content": "Explain HTTP 429 in one sentence."}],
)
print(msg.content[0].text)

import Anthropic from '@anthropic-ai/sdk'

const client = new Anthropic({baseURL: 'https://bigmodel.io', apiKey: process.env.BIGMODEL_API_KEY})

const msg = await client.messages.create({
  model: 'glm-5.3',
  max_tokens: 256,
  system: 'You are a concise assistant.',
  messages: [{role: 'user', content: 'Explain HTTP 429 in one sentence.'}],
})
console.log(msg.content[0].text)

Example response (the id, text and token counts are illustrative):

{
  "id": "msg_...",
  "type": "message",
  "role": "assistant",
  "model": "glm-5.3",
  "content": [
    {"type": "text", "text": "HTTP 429 means you sent too many requests in a given time and should retry later."}
  ],
  "stop_reason": "end_turn",
  "usage": {"input_tokens": 21, "output_tokens": 19, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}
}

#Streaming

with client.messages.stream(
    model="glm-5.3",
    max_tokens=256,
    messages=[{"role": "user", "content": "Count to 3."}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

const stream = client.messages.stream({
  model: 'glm-5.3',
  max_tokens: 256,
  messages: [{role: 'user', content: 'Count to 3.'}],
})
stream.on('text', text => process.stdout.write(text))
await stream.finalMessage()

#Prompt caching

{
  "model": "glm-5.3",
  "max_tokens": 256,
  "system": [
    {"type": "text", "text": "<long, stable instructions>", "cache_control": {"type": "ephemeral"}}
  ],
  "messages": [{"role": "user", "content": "First question"}]
}

#Notes

• Chat-only groups return 404. Keys in DeepSeek or GLM platform groups get not_found_error "… platform only supports /v1/chat/completions; /v1/messages is not available". Use Chat Completions or a key from a group that serves Messages.
• OpenAI groups convert Messages only if the group allows it; otherwise 403 "This group does not allow /v1/messages dispatch".
• model is required; a missing one returns 400 invalid_request_error. An unserved or mistyped model returns 503 api_error "No available accounts", not 404.
• Prompt caching. cache_control passes through. At most 4 breakpoints are kept; extras are removed from tools first, then messages, then system. cache_control on thinking blocks is stripped. Cache writes and reads are billed at the model's cache-write and cache-read prices. See Prompt caching.
• Thinking. thinking and output_config.effort are read and forwarded. If the conversation history contains thinking blocks or tool IDs the model cannot accept, you get 409 malformed_session_state; start a new session. See Reasoning & thinking.
• metadata.user_id. Send a stable value per conversation so its requests are handled consistently, which helps prompt caching.
• Errors use the Anthropic envelope {"type":"error","error":{...}}, except key checks, which use {"code","message"}. Mid-stream errors arrive as an SSE error event. Full list: Errors.

#Related

• Claude Code
• Anthropic SDK
• Count tokens
• Prompt caching