Chat Completions API

curl --request POST \
  --url https://api.minimax.io/v1/chat/completions \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: <content-type>' \
  --data '
{
  "model": "MiniMax-M3.1-Flash-Preview",
  "thinking": {
    "type": "adaptive"
  },
  "messages": [
    {
      "role": "user",
      "content": "Which is bigger, 9.11 or 9.9?"
    }
  ],
  "max_completion_tokens": 500,
  "reasoning_effort": "max"
}
'

{
  "id": "066b366b8e8beeafd3b8b49f89e00648",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "**9.9 is bigger than 9.11.**\n\nIt can be tempting to think 9.11 > 9.9 because \"11\" looks larger than \"9,\" but decimals don't work that way. Here's why:\n\n- **9.11** = 9 + **0.11** (eleven hundredths)\n- **9.9** = 9 + **0.90** (ninety hundredths)\n\nSince 0.90 > 0.11, **9.9 is greater**. ✅\n\n**Tip:** When comparing decimals, compare them to the same number of decimal places. Write 9.9 as 9.90 — now it's easy to see that 9.90 > 9.11.",
        "reasoning_content": "The user is asking which is bigger, 9.11 or 9.9. This is a comparison of two decimal numbers.\n\n9.11 vs 9.9\n\nLet me think carefully:\n- 9.11 = 9 + 0.11\n- 9.9 = 9 + 0.90\n\n0.90 > 0.11, so 9.9 > 9.11.\n\nAlternatively, I can compare them as if they were both decimals to the same place:\n- 9.11\n- 9.90\n\nComparing the tenths place: 1 vs 9. Since 9 > 1, 9.9 is bigger.\n\nSo 9.9 is bigger than 9.11.",
        "role": "assistant",
        "name": "MiniMax AI",
        "audio_content": ""
      }
    }
  ],
  "created": 1780220779,
  "model": "MiniMax-M3.1-Flash-Preview",
  "object": "chat.completion",
  "usage": {
    "total_tokens": 484,
    "total_characters": 0,
    "prompt_tokens": 172,
    "completion_tokens": 312,
    "prompt_tokens_details": {
      "cached_tokens": 159
    }
  },
  "input_sensitive": false,
  "output_sensitive": false,
  "input_sensitive_type": 0,
  "output_sensitive_type": 0,
  "output_sensitive_int": 0,
  "base_resp": {
    "status_code": 0,
    "status_msg": ""
  }
}

POST

chat

completions

curl --request POST \
  --url https://api.minimax.io/v1/chat/completions \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: <content-type>' \
  --data '
{
  "model": "MiniMax-M3.1-Flash-Preview",
  "thinking": {
    "type": "adaptive"
  },
  "messages": [
    {
      "role": "user",
      "content": "Which is bigger, 9.11 or 9.9?"
    }
  ],
  "max_completion_tokens": 500,
  "reasoning_effort": "max"
}
'

{
  "id": "066b366b8e8beeafd3b8b49f89e00648",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "**9.9 is bigger than 9.11.**\n\nIt can be tempting to think 9.11 > 9.9 because \"11\" looks larger than \"9,\" but decimals don't work that way. Here's why:\n\n- **9.11** = 9 + **0.11** (eleven hundredths)\n- **9.9** = 9 + **0.90** (ninety hundredths)\n\nSince 0.90 > 0.11, **9.9 is greater**. ✅\n\n**Tip:** When comparing decimals, compare them to the same number of decimal places. Write 9.9 as 9.90 — now it's easy to see that 9.90 > 9.11.",
        "reasoning_content": "The user is asking which is bigger, 9.11 or 9.9. This is a comparison of two decimal numbers.\n\n9.11 vs 9.9\n\nLet me think carefully:\n- 9.11 = 9 + 0.11\n- 9.9 = 9 + 0.90\n\n0.90 > 0.11, so 9.9 > 9.11.\n\nAlternatively, I can compare them as if they were both decimals to the same place:\n- 9.11\n- 9.90\n\nComparing the tenths place: 1 vs 9. Since 9 > 1, 9.9 is bigger.\n\nSo 9.9 is bigger than 9.11.",
        "role": "assistant",
        "name": "MiniMax AI",
        "audio_content": ""
      }
    }
  ],
  "created": 1780220779,
  "model": "MiniMax-M3.1-Flash-Preview",
  "object": "chat.completion",
  "usage": {
    "total_tokens": 484,
    "total_characters": 0,
    "prompt_tokens": 172,
    "completion_tokens": 312,
    "prompt_tokens_details": {
      "cached_tokens": 159
    }
  },
  "input_sensitive": false,
  "output_sensitive": false,
  "input_sensitive_type": 0,
  "output_sensitive_type": 0,
  "output_sensitive_int": 0,
  "base_resp": {
    "status_code": 0,
    "status_msg": ""
  }
}

✨ New model — MiniMax-M3.1-Flash-PreviewCore capabilities: 1M long context, multimodal, tunable thinking depth.Available only through M Plan and MiniMax Code for now.

What’s new in MiniMax-M3.1-Flash-Preview:

Image and video understanding — see the example code on the right
Tune thinking depth via reasoning_effort, which accepts low, medium, high, xhigh, and max; the default is max
Thinking is on by default with no configuration needed; thinking content is returned separately on the reasoning_content field

Authorizations

Authorization

string

header

required

HTTP: Bearer Auth

Security Scheme Type: http
HTTP Authorization Scheme: Bearer API_key, used for account verification, can be viewed in Account Management > API Keys

Headers

Content-Type

enum<string>

default:application/json

required

Media type of the request body, should be set to application/json to ensure JSON format

Available options:

application/json

Body

application/json

model

enum<string>

required

Model ID

Available options:

MiniMax-M3.1-Flash-Preview,

MiniMax-M3,

MiniMax-M2.7,

MiniMax-M2.7-highspeed,

MiniMax-M2.5,

MiniMax-M2.5-highspeed,

MiniMax-M2.1,

MiniMax-M2.1-highspeed,

MiniMax-M2

messages

object[]

required

A list of messages containing the conversation history. Supports text, image, video, and tool call content.

Show child attributes

service_tier

enum<string>

default:standard

Service tier for request admission. Supported values are standard and priority. If omitted, the request uses the standard tier. The priority price is 1.5 times the standard price and ensures priority admission so the request is processed ahead of other requests, leading to faster responses and fewer failures.

Available options:

standard,

priority

thinking

object

Controls thinking behavior. The adaptive default applies to every model here; only the effect of disabled differs.

MiniMax-M3.1-Flash-Preview: always thinks. disabled returns HTTP 400. Use reasoning_effort to tune thinking depth.
MiniMax-M3: adaptive thinking is on by default; disabled skips thinking and answers directly.
M2.x models: thinking cannot be disabled; disabled is accepted but ignored.

Show child attributes

reasoning_effort

enum<string>

Tunes thinking depth. Only MiniMax-M3.1-Flash-Preview supports real thinking-depth tuning; other models ignore this field. When omitted, its default thinking depth is max. none is not accepted and returns HTTP 400 because MiniMax-M3.1-Flash-Preview always thinks.

Available options:

low,

medium,

high,

xhigh,

max

reasoning_split

boolean

Output-format switch. When true, thinking content is separated into the reasoning_content field; when false, thinking is kept inside content wrapped in <think> tags. MiniMax-M3.1-Flash-Preview does not currently support setting this parameter to false. This parameter does not enable or disable thinking.

stream

boolean

default:false

Whether to use streaming output, defaults to false. When set to true, the response will be returned in chunks.

stream_options

object

Streaming response options.

Show child attributes

max_completion_tokens

integer<int64>

Specifies the upper limit for generated content length, in tokens. For MiniMax-M3.1-Flash-Preview and MiniMax-M3 the recommended value is 131072 (128K) and the maximum is 524288 (512K); for other models the recommended value is 65536 (64K) and the maximum is 204800 (200K). If generation stops due to length, try increasing this value.

Required range: x >= 1

temperature

number<double>

default:1

Temperature coefficient, affects output randomness. Range [0, 2], default 1. Higher values produce more random output; lower values produce more deterministic output.

Required range: 0 <= x <= 2

top_p

number<double>

default:0.95

Nucleus sampling parameter. Range [0, 1]. Default is 0.95 for MiniMax-M3.1-Flash-Preview and MiniMax-M3, and 0.9 for M2.x models.

Required range: 0 <= x <= 1

tools

object[]

Tool definition list. Function tools are supported.

Show child attributes

max_tokens

integer<int64>

deprecated

Legacy generation length limit parameter. Deprecated; use max_completion_tokens instead.

Required range: x >= 1

Response

string

Unique ID of this response

choices

object[]

List of response choices

Show child attributes

created

integer<int64>

Unix timestamp (seconds) when the response was created

model

string

Model ID used for this request

object

enum<string>

Object type. chat.completion for non-streaming, chat.completion.chunk for streaming

Available options:

chat.completion,

chat.completion.chunk

usage

object

Token usage statistics for this request

Show child attributes

input_sensitive

boolean

Whether the input content triggered sensitive word detection. If the input content is severely inappropriate, the API will return a content violation error message with empty reply content

input_sensitive_type

integer<int64>

Type of sensitive word triggered by input, returned when input_sensitive is true. Values: 1 Severe violation; 2 Pornography; 3 Advertising; 4 Prohibited; 5 Abuse; 6 Violence/Terrorism; 7 Other

output_sensitive

boolean

Whether the output content triggered sensitive word detection. If the output content is severely inappropriate, the API will return a content violation error message with empty reply content

output_sensitive_type

integer<int64>

Type of sensitive word triggered by output

base_resp

object

Error status code and details

Show child attributes

Using the API

Text

Video

Speech

Image

Music

File

Models

Authorizations

Headers

Body

Response