MaaS_MiniMax_M3.1

Model Overview

Item	value
Model ID (model)	MiniMax-M3.1-Flash
context window	1,000,000 tokens (total of input and output)
Maximum Output	For the max_completion_tokensparameter of Chat Completions and the max_tokensparameter of Anthropic Messages: the recommended value is 131072 (128K), the upper limit is 524288 (512K), and the minimum value is 1. The max_output_tokensparameter of Responses does not have the same specified values
input modality	text, images, and videos
Output Modality	Text; additionally returns reasoning content and supports tool calls
Default Thinking Mode	Enabled and cannot be disabled. The default thinking depth is max, and can be adjusted to low, medium, high, xhigh, max
Explicit Prompt Cache	Anthropic cache_controlis not supported. The response usagemay still return cache counts; Responses provide the prompt_cache_key

Request Protocol

https

Header

Parameter Name	Type	Required	Description
Content-Type	string	Yes	application/json
Authorization	string	Yes	Bearer ${your_AK}

OpenAI Chat Completions

Request URL

POST https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions

Request Body Parameters

Parameter Name	Secondary Parameters	Three-level parameters	CET-4 Parameters	Type	Required	Description
model	-	-	-	string	Yes	Fixed as MaaS_Minimax_M3.1_flash_20261019 (original model from the vendor: MiniMax-M3.1-Flash)
messages	-	-	-	array[object]	Yes	Conversation history. Supports text, images, videos and tool calls. For multi-round tool calls, the complete assistant message (including tool_calls) and reasoning_content must be carried back as is.
	role	-	-	string	Yes	system, user, assistant, tool
	name	-	-	string	No	Sender name. Used for differentiation when multiple instances of the same role exist.
	content	-	-	string or array	No	An array of text or content blocks. The schema only lists roleas required
		type	-	string	Yes	Required when the content block is an array: text, image_url, video_url
		text	-	string	No	type= texttext at this time
		image_url	-	object	No	type= image_url. Supported formats: JPEG, PNG, GIF, WEBP. Single image file size ≤ 10 MB, total request body size ≤ 64 MB
			url	string	Yes	Image URL or Base64 data URL
			detail	string	No	low, default, high, default value is default. Approximate per-image token usage: low usually ranges from several hundred up to around 600 tokens; default typically falls between 1k–3k, with a maximum of roughly 5k; high usually sits in the thousands, topping out at 15k+. Refer to the response usagefor the exact figure
			max_long_side_pixel	integer	No	Longest side in pixels, minimum 1
		video_url	-	object	No	type= video_url. Supported formats: MP4, AVI, MOV, MKV. URL/Base64 video size ≤ 50 MB;
			url	string	Yes	Video URL or Base64 data URL
			detail	string	No	Frame extraction resolution: low, default, high, default is default
			fps	number	No	Frame extraction frequency, default value is 1, range [0.2, 5]. A higher value means higher sensitivity, more tokens and slower processing speed.
			max_long_side_pixel	integer	No	Maximum side pixel count of a single frame, minimum value: 1
	tool_calls	-	-	array[object]	No	Tool calls in the assistant message. Must be echoed back exactly as is in the next round
		id	-	string	Yes	Tool Call ID
		type	-	string	Yes	Fixed function
		function	-	object	Yes	function call
			name	string	Yes	function name
			arguments	string	Yes	function input parameter, JSON string
	tool_call_id	-	-	string	No	Required when role= tool, take the value of tool_calls[]. idfrom the previous round. Ignored for other roles
thinking	-	-	-	object	No	This model enforces thinking; if omitted, it is equivalent to being enabled. disabledReturns HTTP 400 (error code 2013)
	type	-	-	string	No	This model only accepts adaptive, with adaptiveas the default. disabledwill return HTTP 400
reasoning_effort	-	-	-	string	No	Depth of thinking, only this model actually takes effect. low, medium, high, xhigh, max, default value maxis omitted. nonereturns HTTP 400. The higher the gear, the more thinking tokens and the higher the latency.
reasoning_split	-	-	-	boolean	No	true, the thinking content goes into reasoning_content; when set to false, the thinking content remains in contentvia think tags. This model does not support false. This field does not toggle the thinking function. No default value is provided in the source document.
stream	-	-	-	boolean	No	Whether to use streaming; defaults to false
stream_options	-	-	-	object	No	Streaming Options
	include_usage	-	-	boolean	No	Whether to return token usage in the stream, default false
max_completion_tokens	-	-	-	integer	No	Maximum generation length. This model recommends 131072 (128K), with a maximum limit of 524288 (512K) and a minimum value of 1. When generation stops due to length, finish_reasonwill be length
temperature	-	-	-	number	No	Sampling temperature, range [0,2], default 1
top_p	-	-	-	number	No	Nuclear sampling, range [0,1]. The default value for this model is 0.95
tools	-	-	-	array[object]	No	Tool definition, currently only supports function
	type	-	-	string	Yes	Fixed function
	function	-	-	object	Yes	function definition
		name	-	string	Yes	function name
		description	-	string	No	function description
		parameters	-	object	Yes	function parameter JSON Schema
max_tokens	-	-	-	integer	No	Deprecated, please use max_completion_tokensinstead. Minimum value is 1. Thinking tokens count towards this limit; a value too small will result in finish_reason= lengthand an empty content.

Request Example

curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer ${your_AK}' \
  -d '{
    "model": "MaaS_Minimax_M3.1_flash_20261019",
    "reasoning_effort": "max",
    "thinking": {"type": "adaptive"},
    "max_completion_tokens": 131072,
    "messages": [
      {"role": "user", "content": "Which is bigger, 9.11 or 9.9?"}
    ]
  }'

Response Example

{
  "id": "066b36619b147e326d17053cccdef70f",
  "object": "chat.completion",
  "created": 1780220769,
  "model": "MaaS_Minimax_M3.1_flash_20261019",
  "choices": [
    {
      "index": 0,
      "finish_reason": "stop",
      "message": {
        "role": "assistant",
        "content": "9.9 is bigger. 9.9 = 9.90, which is greater than 9.11.",
        "reasoning_content": "Comparing 9.11 and 9.90, the decimal part 90 > 11, so 9.9 is bigger."
      }
    }
  ],
  "usage": {
    "prompt_tokens": 170,
    "completion_tokens": 303,
    "total_tokens": 473,
    "prompt_tokens_details": {"cached_tokens": 157}
  },
  "input_sensitive": false,
  "output_sensitive": false,
  "base_resp": {"status_code": 0, "status_msg": ""}
}

finish_reasonValue options: stop, length, content_filter, tool_calls. Simple turns may not return reasoning_content, so check for null before reading. The Usage field in OpenAPI only declares total_tokens; official examples also return prompt_tokens, completion_tokens, and prompt_tokens_details.cached_tokens.

Example of streaming request

curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer ${your_AK}' \
  -d '{
    "model": "MaaS_Minimax_M3.1_flash_20261019",
    "reasoning_effort": "max",
    "stream": true,
    "stream_options": {"include_usage": true},
    "messages": [{"role": "user", "content": "Which is bigger, 9.11 or 9.9?"}]
  }'

```plain text data: {"id":"066a2db7cb70134c45f5d6443d434c2c","object":"chat.completion.chunk","created":1780153015,"model":"MaaS_Minimax_M3.1_flash_20261019","choices":[{"index":0,"delta":{"role":"assistant","reasoning_content":"Comparing the two decimal numbers"},"finish_reason":null}]}

data: {"id":"066a2db7cb70134c45f5d6443d434c2c","object":"chat.completion.chunk","created":1780153015,"model":"MaaS_Minimax_M3.1_flash_20261019","choices":[{"index":0,"delta":{"role":"assistant","content":"9.9 is bigger."},"finish_reason":"stop"}]}

`The finish_reason of streaming chunks in the origin server enumeration is stop`, `length`, and is null when not finished.

### SDK Calling Method

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://genaiapi.cloudsway.net/v1/ai/{endpointPath}",
    api_key="${your_AK}",
)

response = client.chat.completions.create(
    model="MaaS_Minimax_M3.1_flash_20261019",
    reasoning_effort="max",
    messages=[{"role": "user", "content": "Which is bigger, 9.11 or 9.9?"}],
    extra_body={"thinking": {"type": "adaptive"}},
)
print(response.choices[0].message.reasoning_content)
print(response.choices[0].message.content)

OpenAI Responses

Request URL

POST https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses

Request Body Parameters

Parameter Name	Secondary Parameters	Three-level parameters	CET-4 Parameters	Type	Required	Description
model	-	-	-	string	Yes	Fixed as MaaS_Minimax_M3.1_flash_20261019 (original model from the vendor: MiniMax-M3.1-Flash)
input	-	-	-	string or array	Yes	Plain text, or an array of conversation history
	type	-	-	string	No	Array item types: message (default), function_call, function_call_output, reasoning
	role	-	-	string	No	When type=message: user, assistant, system, developer, tool
	content	-	-	string or array	No	type= messagetext or multimodal content blocks
		type	-	string	Yes	input_text, output_text, input_image, input_video
		text	-	string	No	input_text/ output_texttext
		image_url	-	string or object	No	input_image. Formats: JPEG, PNG, GIF, WEBP
			url	string	Yes	Required when in object format; valid values: image URL or Base64 string
			detail	string	No	low, default, high; the default is default
		video_url	-	string or object	No	input_video. Formats: MP4, AVI, MOV, MKV. Large files are processed via the File API
			url	string	Yes	Required when in object format; valid values: video URL or Base64
			fps	number	No	Frame extraction frequency, default value 1, range [0.2, 5]
			detail	string	No	low, default, high; the default is default
			max_long_side_pixel	integer	No	The pixel count of the longest edge of the video frame. No minimum/maximum values are provided in the source document.
	call_id	-	-	string	No	function_callor function_call_outputtool call ID
	name	-	-	string	No	function_callfunction name
	arguments	-	-	string	No	function_callparameters, JSON string
	output	-	-	string or array	No	function_call_outputtool response
	summary	-	-	array	No	type= reasoningreasoning snippet
		type	-	string	No	summary_text
		text	-	string	No	Reasoning Text
instructions	-	-	-	string	No	System Instructions
max_output_tokens	-	-	-	integer	No	Maximum output tokens. Thought tokens count towards this limit; a value that is too small will result in status= incomplete, and there will be no messagein output. The source document does not specify the upper and lower limits for this value
temperature	-	-	-	number	No	Sampling temperature. The specified range is (0,1], with a default value of 1; the schema defines a minimum of 0 and a maximum of 1. This differs from the [0,2] range used for Chat Completions
top_p	-	-	-	number	No	Kernel sampling. The description is in the range (0,1], with a default value of 0.95; the minimum of the schema is 0 and the maximum is 1
stream	-	-	-	boolean	No	truemeans moving SSE when enabled, default value is false
tools	-	-	-	array[object]	No	Tool List
	type	-	-	string	Yes	Fixed function
	name	-	-	string	Yes	function name
	description	-	-	string	No	function description
	parameters	-	-	object	No	Parameter JSON Schema
tool_choice	-	-	-	string	No	Only auto, where the model decides whether to invoke tools
metadata	-	-	-	object	No	Request metadata, where both keys and values are strings
prompt_cache_key	-	-	-	string	No	Prompt cache routing identifier
text	-	-	-	object	No	Output Format Control
	format	-	-	object	No	format object
		type	-	string	No	Only text, default text
reasoning	-	-	-	object	No	This model always performs reasoning even when reasoningis omitted
	effort	-	-	string	No	The model-specific clause accepts low, medium, high, xhigh, max; the default value maxmay be omitted. nonereturns HTTP 400. The schema enumeration additionally includes minimal, which the source document does not specify whether the model accepts.

Request Example

curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer ${your_AK}' \
  -d '{
    "model": "MaaS_Minimax_M3.1_flash_20261019",
    "reasoning": {"effort": "max"},
    "input": "Which is bigger, 9.11 or 9.9?"
  }'

Response Example

{
  "id": "07229fa48d63ee86b664bb1c3a2257b7",
  "object": "response",
  "created_at": 1792240804,
  "model": "MaaS_Minimax_M3.1_flash_20261019",
  "status": "completed",
  "output": [
    {
      "id": "07229fa48d63ee86b664bb1c3a2257b7_rs",
      "type": "reasoning",
      "status": "completed",
      "summary": [],
      "content": [{"type": "reasoning_text", "text": "9.9 = 9.90, which is greater than 9.11."}]
    },
    {
      "id": "07229fa48d63ee86b664bb1c3a2257b7_msg",
      "type": "message",
      "status": "completed",
      "role": "assistant",
      "content": [{"type": "output_text", "text": "9.9 is bigger than 9.11.", "annotations": []}]
    }
  ],
  "output_text": "9.9 is bigger than 9.11.",
  "usage": {
    "input_tokens": 218,
    "output_tokens": 68,
    "total_tokens": 286,
    "input_tokens_details": {"cached_tokens": 189},
    "output_tokens_details": {"reasoning_tokens": 36}
  },
  "error": null,
  "incomplete_details": null
}

status values: completed, incomplete, failed. incomplete_details.reason values: max_output_tokens, content_filter. output may also contain type=function_call (including call_id, name, arguments).

Stream Request Example

curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer ${your_AK}' \
  -d '{
    "model": "MaaS_Minimax_M3.1_flash_20261019",
    "reasoning": {"effort": "max"},
    "stream": true,
    "input": "Which is bigger, 9.11 or 9.9?"
  }'

The streaming event types from the origin server, in order, include: response.created, response.in_progress, response.output_item.added, response.content_part.added, response.reasoning_text.delta, response.reasoning_text.done, response.content_part.done, response.output_item.done, response.output_text.delta, response.output_text.done, response.completed.

data: {"type":"response.created","sequence_number":0,"response":{"id":"07229fa484712ce19b28c3e706b4f60c","object":"response","model":"MaaS_Minimax_M3.1_flash_20261019","status":"in_progress","output":[]}}

data: {"type":"response.reasoning_text.delta","sequence_number":5,"delta":"9.9 = 9.90"}

data: {"type":"response.output_text.delta","sequence_number":12,"delta":"9.9 is bigger."}

data: {"type":"response.completed","sequence_number":20,"response":{"id":"07229fa484712ce19b28c3e706b4f60c","object":"response","model":"MaaS_Minimax_M3.1_flash_20261019","status":"completed"}}

SDK Calling Method

from openai import OpenAI

client = OpenAI(
    base_url="https://genaiapi.cloudsway.net/v1/ai/{endpoint}",
    api_key="${your_AK}",
)

response = client.responses.create(
    model="MaaS_Minimax_M3.1_flash_20261019",
    reasoning={"effort": "max"},
    input="Which is bigger, 9.11 or 9.9?",
)
print(response.output_text)

Anthropic Messages

Request URL

POST https://genaiapi.cloudsway.net/{endpoint}/v1/messages

Request Body Parameters

Parameter Name	Secondary Parameters	Three-level parameters	CET-4 Parameters	Type	Required	Description
model	-	-	-	string	Yes	Fixed as MaaS_Minimax_M3.1_flash_20261019 (original model from the vendor: MiniMax-M3.1-Flash)
max_tokens	-	-	-	integer	No	Not listed as required in the OpenAPI, but all examples will pass a value. This model recommends 131072 (128K), with a maximum limit of 524288 (512K) and a minimum value of 1. The reasoning tokens count towards this upper limit; an excessively small value will result in stop_reason= max_tokenswith no text output.
messages	-	-	-	array[object]	Yes	Dialogue history. The model uses alternating user/ assistantformat. Subsequent turns must carry back the thinkingblock (including the signature) as is
	role	-	-	string	Yes	This model uses user, assistant. The schema also lists user_system, group, sample_message_user, sample_message_ai, and the source document does not specify whether this model supports them.
	content	-	-	string or array	Yes	Plain text, or an array of content blocks
		type	-	string	Yes	text, image, video, tool_use, tool_result, thinking, mid_conv_system. thinkingis only used for postback of the previous round's assistant thinking
		text	-	string	No	type= texttext
		source	-	object	No	type= imageor type= videomedia source
			type	string	Yes	base64or url
			media_type	string	No	Required when using base64, e. g., image/png, video/mp4. Image: JPEG/PNG/GIF/WEBP. Video: MP4, AVI, MOV, MKV
			data	string	No	Base64 Media Bytes
			url	string	No	Public network URL. URL/Base64: video ≤ 50 MB, image ≤ 10 MB, request body ≤ 64 MB;
			detail	string	No	low, default, high, default is default. The approximate image usage is the same as the detail table for Chat Completions.
			fps	number	No	Video frame extraction, default value is 1, range is [0.2, 5]
			max_long_side_pixel	integer	No	The pixel count of the longest edge of an image or video frame, minimum value: 1
		id	-	string	No	postback tool_usetool call ID
		name	-	string	No	postback tool_usetool name
		input	-	object	No	postback tool_usetool input parameters
		tool_use_id	-	string	No	The tool_use ID corresponding to tool_result
		content	-	string or array	No	tool_resultis the execution result, which can be a string or an array of text/image content blocks
		thinking	-	string	No	postback thought text
		signature	-	string	No	Consider the signature, which must be postedback as-is in subsequent rounds
system	-	-	-	string or array	No	System prompt. Array items are type= textplus text. cache_controlnot supported by this model
stream	-	-	-	boolean	No	Whether to use streaming; defaults to false
temperature	-	-	-	number	No	Range [0,2], default value is 1. The recommended value for compatibility note is 1
top_p	-	-	-	number	No	Range [0,1]. The default value of this model is 0.95
tools	-	-	-	array[object]	No	Tool Definition
	name	-	-	string	Yes	Tool Name
	description	-	-	string	No	Tool Description
	input_schema	-	-	object	Yes	Input JSON Schema
tool_choice	-	-	-	object	No	Tool selection strategy, only supports auto
	type	-	-	string	Yes	Fixed auto
thinking	-	-	-	object	No	This model is always active. If you pass type, it can only be adaptive;disabledwill return HTTP 400
	type	-	-	string	No	This model only accepts adaptive. disabledreturns HTTP 400
output_config	-	-	-	object	No	Output configuration. Use effortto adjust the thinking depth of this model
	effort	-	-	string	No	low, medium, high, xhigh, max; when omitted, the default is max. none returns HTTP 400
metadata	-	-	-	object	No	Request metadata
	user_id	-	-	string	No	End-user ID, used for aggregation, rate limiting and billing analysis

Request Example

curl -X POST 'https://genaiapi.cloudsway.net/{endpoint}/v1/messages' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer ${your_AK}' \
  -d '{
    "model": "MaaS_Minimax_M3.1_flash_20261019",
    "max_tokens": 4096,
    "thinking": {"type": "adaptive"},
    "output_config": {"effort": "max"},
    "messages": [
      {"role": "user", "content": "Which is bigger, 9.11 or 9.9?"}
    ]
  }'

Response Example

{
  "id": "066b367ba949a58a9d6829b4bf8cbcd4",
  "type": "message",
  "role": "assistant",
  "model": "MaaS_Minimax_M3.1_flash_20261019",
  "content": [
    {
      "type": "thinking",
      "thinking": "9.9 = 9.90, which is greater than 9.11.",
      "signature": "6d0315c818f9664ff185dabaa22cd89f2bf28a3a52122095bce23d905471ec5f"
    },
    {
      "type": "text",
      "text": "9.9 is bigger than 9.11."
    }
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 13,
    "output_tokens": 149,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 159
  }
}

stop_reasonValue options: end_turn, max_tokens, tool_use. When a tool is called, the content block is tool_use (id, name, input). The response will not contain image, video, tool_result, or mid_conv_system blocks.

Stream Request Example

curl -X POST 'https://genaiapi.cloudsway.net/{endpoint}/v1/messages' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer ${your_AK}' \
  -d '{
    "model": "MaaS_Minimax_M3.1_flash_20261019",
    "max_tokens": 4096,
    "output_config": {"effort": "max"},
    "stream": true,
    "messages": [{"role": "user", "content": "Which is bigger, 9.11 or 9.9?"}]
  }'

Event types: message_start, ping, content_block_start, content_block_delta, content_block_stop, message_delta, message_stop. The delta types include thinking_delta, text_delta, signature_delta. When an error occurs during streaming, it is event: error.

event: message_start
data: {"type":"message_start","message":{"id":"066b367ba949a58a9d6829b4bf8cbcd4","type":"message","role":"assistant","model":"MaaS_Minimax_M3.1_flash_20261019","content":[],"stop_reason":null}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":"9.9 = 9.90"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"9.9 is bigger."}}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":149}}

event: message_stop
data: {"type":"message_stop"}

SDK Calling Method

import anthropic

client = anthropic.Anthropic(
    base_url="https://genaiapi.cloudsway.net/{endpoint}/v1",
    api_key="${your_AK}",
)

message = client.messages.create(
    model="MaaS_Minimax_M3.1_flash_20261019",
    max_tokens=4096,
    output_config={"effort": "max"},
    messages=[{"role": "user", "content": "Which is bigger, 9.11 or 9.9?"}],
)

for block in message.content:
    if block.type == "thinking":
        print(block.thinking)
    elif block.type == "text":
        print(block.text)