curl --request POST \
--url https://direct.evolink.ai/v1/messages \
--header 'Authorization: Bearer <token>' \
--header 'Content-Type: application/json' \
--data '
{
"model": "MiniMax-M3",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello, world"
}
]
}
'{
"id": "066a381bdc3c0ded310e27c9a46d16e7",
"type": "message",
"role": "assistant",
"model": "MiniMax-M3",
"content": [
{
"type": "thinking",
"thinking": "The user is asking about the capital of Japan, which is a basic geography question. The answer is Tokyo, I can give it directly.",
"signature": "066a381bdc3c0ded310e27c9a46d16e7"
},
{
"type": "text",
"text": "The capital of Japan is **Tokyo**."
}
],
"stop_reason": "end_turn",
"usage": {
"input_tokens": 7,
"output_tokens": 77,
"cache_creation_input_tokens": 0,
"cache_read_input_tokens": 0
}
}MiniMax-M3 - Anthropic-Compatible API
- Use the Anthropic Messages protocol to call the MiniMax-M3 model
- Request / response structure aligns with the Anthropic API
- Multimodal conversation:
contentsupports text and image content blocks - System prompts: Passed via the top-level
systemfield - Thinking mode: Controlled via the
thinkingobject; thinking content is returned viacontent[type=thinking]block - Streaming output: SSE event stream
- Tool calling: Compatible with Anthropic
tool_use/tool_resultflow
curl --request POST \
--url https://direct.evolink.ai/v1/messages \
--header 'Authorization: Bearer <token>' \
--header 'Content-Type: application/json' \
--data '
{
"model": "MiniMax-M3",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello, world"
}
]
}
'{
"id": "066a381bdc3c0ded310e27c9a46d16e7",
"type": "message",
"role": "assistant",
"model": "MiniMax-M3",
"content": [
{
"type": "thinking",
"thinking": "The user is asking about the capital of Japan, which is a basic geography question. The answer is Tokyo, I can give it directly.",
"signature": "066a381bdc3c0ded310e27c9a46d16e7"
},
{
"type": "text",
"text": "The capital of Japan is **Tokyo**."
}
],
"stop_reason": "end_turn",
"usage": {
"input_tokens": 7,
"output_tokens": 77,
"cache_creation_input_tokens": 0,
"cache_read_input_tokens": 0
}
}https://direct.evolink.ai, which has better support for text models and long-lived connections. https://api.evolink.ai is the primary endpoint for multimodal services and serves as a fallback address for text models.Authorizations
##All APIs require Bearer Token authentication##
Get API Key:
Visit API Key Management Page to get your API Key
Add to request header:
Authorization: Bearer YOUR_API_KEY
Note: EvoLink uses Bearer Token authentication uniformly for /v1/messages.
Body
Model to call
MiniMax-M3 "MiniMax-M3"
List of conversation messages, alternating user / assistant turns
Notes:
- Must contain at least 1 message
- The last message is typically
role=user
1Show child attributes
Show child attributes
Upper limit for generated content length (in tokens)
Notes:
- MiniMax-M3 recommended 131,072 (128K), maximum 524,288 (512K)
- Tokens generated by thinking also count toward this limit
- Content exceeding the limit will be truncated; if generation is interrupted due to
length, try increasing this value
1 <= x <= 5242881024
System prompt, used to set the AI's role and behavior
Notes:
- Supports a string or an array of strings
- Passed via the top-level
systemfield
"You are a helpful assistant."
Sampling temperature
Notes:
- Range:
[0, 2] - Default 1; higher values produce more divergent output, lower values produce more deterministic output
0 <= x <= 21
Nucleus sampling threshold
Notes:
- Range:
[0, 1], MiniMax-M3 default 0.95 - It is recommended not to adjust temperature and top_p simultaneously
0 <= x <= 10.95
Whether to return via SSE streaming
true: Server-Sent Events streaming responsefalse: Wait for complete response before returning (default)
false
Controls deep thinking. When thinking is enabled, thinking blocks must be passed back as-is in multi-turn conversations
Notes:
- Defaults to
adaptive: The model adaptively decides whether to engage in deep thinking based on problem difficulty - When enabled, the response
contentarray will include atype="thinking"reasoning block (billed as output tokens)
Show child attributes
Show child attributes
Tool definition list
Notes:
- Follows the Anthropic tool definition specification
input_schemauses a JSON Schema object
Show child attributes
Show child attributes
Tool selection strategy. Only auto and none are supported
Show child attributes
Show child attributes
Request metadata
Show child attributes
Show child attributes
Response
Message object
Anthropic-style message response
Unique message ID
Response object type
message assistant Model actually used
"MiniMax-M3"
Response content block list
Possible block types:
thinking: Reasoning process (only when thinking is active)text: Final answer texttool_use: Tool call initiated by the model
Show child attributes
Show child attributes
Stop reason
end_turn: Natural completionmax_tokens: Reached max_tokens limittool_use: Model triggered a tool call
end_turn, max_tokens, tool_use Token usage statistics (Anthropic specification)
Show child attributes
Show child attributes