> ## Documentation Index
> Fetch the complete documentation index at: https://platform.minimax.io/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Messages API
> Use the Anthropic API compatible Messages format to call MiniMax models.
✨ **New model — `MiniMax-M3`**
**Core capabilities**: **Coding/Agentic SOTA**, **1M long context**, **multimodal**.
**What's new in `MiniMax-M3`:**
1. Image and video understanding — see the example code on the right
2. Control thinking via the `thinking` parameter
## OpenAPI
````yaml /api-reference/text/api/openapi-chat-anthropic.json POST /anthropic/v1/messages
openapi: 3.1.0
info:
title: MiniMax Text API Anthropic
description: >
MiniMax text generation API with support for chat completion and streaming
output.
> ⚡ **`MiniMax-M3` highlights** — **Coding/Agentic SOTA**, **1M long
context**, and **multimodal**.
license:
name: MIT
version: 1.0.0
servers:
- url: https://api.minimax.io
security:
- bearerAuth: []
- apiKeyAuth: []
paths:
/anthropic/v1/messages:
post:
tags:
- Text Generation
summary: Text Generation Anthropic
operationId: chatCompletionAnthropic
parameters:
- name: Content-Type
in: header
required: true
description: >-
Media type of the request body, should be set to `application/json`
to ensure JSON format
schema:
type: string
enum:
- application/json
default: application/json
requestBody:
content:
application/json:
schema:
$ref: '#/components/schemas/CreateMessageReq'
examples:
Image Understanding:
value:
model: MiniMax-M3
thinking:
type: adaptive
messages:
- role: user
content:
- type: text
text: What does this image show?
- type: image
source:
type: url
url: >-
https://filecdn.minimax.chat/public/fe9d04da-f60e-444d-a2e0-18ae743add33.jpeg
max_tokens: 500
Video Understanding:
value:
model: MiniMax-M3
thinking:
type: adaptive
messages:
- role: user
content:
- type: text
text: What is happening in this video?
- type: video
source:
type: url
url: >-
https://filecdn.minimax.chat/public/ee8c1648-21f1-41b7-8397-65022d22ffe5.mp4
max_tokens: 1024
Deep Thinking:
value:
model: MiniMax-M3
thinking:
type: adaptive
messages:
- role: user
content: Which is bigger, 9.11 or 9.9?
max_tokens: 1024
Stream:
value:
model: MiniMax-M3
thinking:
type: adaptive
messages:
- role: user
content:
- type: text
text: What does this image show?
- type: image
source:
type: url
url: >-
https://filecdn.minimax.chat/public/fe9d04da-f60e-444d-a2e0-18ae743add33.jpeg
stream: true
max_tokens: 500
Tool Use:
value:
model: MiniMax-M3
messages:
- role: user
content: How's the weather in San Francisco?
max_tokens: 1024
tools:
- name: get_weather
description: Get the current weather for a given location.
input_schema:
type: object
properties:
location:
type: string
description: The city and state/country, e.g. San Francisco, US
required:
- location
tool_choice:
type: auto
required: true
responses:
'200':
description: ''
content:
application/json:
schema:
$ref: '#/components/schemas/CreateMessageResp'
examples:
Image Understanding:
value:
id: 066a3730f92943410551c4b4978ddad1
type: message
role: assistant
model: MiniMax-M3
content:
- thinking: >-
The user is asking what the image shows. I can
describe this image of a young child.
signature: >-
db8ecae1e5a63cabebd69d2625983499bb498bfd329dde5c3cede76f12896336
type: thinking
- text: >-
This image shows a young girl, probably around 4-5
years old, photographed in a close-up portrait style.
She has:
- **Wavy, tousled light brown hair** with bangs that
fall across her forehead, with some strands appearing
slightly messy or windblown
- **Large, expressive hazel-green eyes** that are
looking directly at the camera
- **A soft, subtle smile** giving her a warm, gentle
expression
- **Fair skin** with rosy cheeks, illuminated by warm,
soft lighting
She's wearing what appears to be a **cream or
off-white dress** with delicate lace or ruffled
detailing around the neckline and shoulders.
The photograph has a **professional portrait
quality**, with a softly blurred, neutral
grayish-beige background and warm directional lighting
that creates a classic, painterly aesthetic. The
composition and styling suggest this could be a formal
portrait or an AI-generated image with a
vintage/realistic photographic style.
type: text
usage:
input_tokens: 1209
output_tokens: 214
cache_creation_input_tokens: 0
cache_read_input_tokens: 157
stop_reason: end_turn
Video Understanding:
value:
id: 066b458956dd2902de31eae19cf15c6e
type: message
role: assistant
model: MiniMax-M3
content:
- thinking: >-
The user wants to know what is happening in the video.
1. **Analyze the video:**
* The video is a meme/edited clip.
* It starts with a cat (specifically a Bengal cat, looks like a Savannah cat or similar breed) walking on a diving board at the Olympics (London 2012).
* The cat is edited into the scene, replacing a human diver.
* The cat walks to the end of the board.
* The cat then jumps off the diving board into the pool.
* The people in the background are real, reacting (or rather, just sitting there, as it's a static crowd image or very slow movement).
* The text "MINIMAX | Hailuo AI" is visible, indicating it's an AI-generated video.
2. **Describe the action:**
* A cat walks on a diving board.
* The cat jumps off the diving board.
* The setting is an Olympic-sized pool.
3. **Synthesize the description:**
* Start with the main subject: a cat.
* Describe the location: a diving board at a swimming pool.
* Describe the action: walking, then jumping off.
* Combine into a concise sentence: "A cat is seen walking on a diving board and then jumping off into a swimming pool."
4. **Review against the video:**
* Does the video show a cat? Yes.
* Does it show a diving board? Yes.
* Does the cat jump off? Yes.
* Is there a swimming pool below? Yes.
* The description is accurate and covers the key events in the short clip.
5. **Final Output Generation:** Present the
synthesized description clearly.
signature: >-
4849e0d596e25f03504dbb4eee7316fc48714dc86b93c9b7dd63adef7f4aa9b2
type: thinking
- text: >-
A cat is seen walking on a diving board and then
jumping off into a swimming pool.
type: text
usage:
input_tokens: 349
output_tokens: 393
cache_creation_input_tokens: 0
cache_read_input_tokens: 1896
stop_reason: end_turn
Deep Thinking:
value:
id: 066b367ba949a58a9d6829b4bf8cbcd4
type: message
role: assistant
model: MiniMax-M3
content:
- thinking: |-
The user is asking which is bigger, 9.11 or 9.9.
9.11 vs 9.9
9.9 is greater than 9.11.
To compare: 9.11 = 9.11 and 9.9 = 9.90
9.90 > 9.11, so 9.9 is bigger.
signature: >-
6d0315c818f9664ff185dabaa22cd89f2bf28a3a52122095bce23d905471ec5f
type: thinking
- text: >-
**9.9 is bigger than 9.11.**
To compare decimals, it helps to write them with the
same number of decimal places:
- 9.11 = 9.11
- 9.9 = 9.90
Since 9.90 > 9.11, **9.9 is larger**.
type: text
usage:
input_tokens: 13
output_tokens: 149
cache_creation_input_tokens: 0
cache_read_input_tokens: 159
stop_reason: end_turn
Stream:
value:
- type: message_start
message:
id: 066a3766d21ce23c8afc5e85a427036c
type: message
role: assistant
content: []
model: MiniMax-M3
stop_reason: null
stop_sequence: null
usage:
input_tokens: 0
output_tokens: 0
cache_creation_input_tokens: 0
cache_read_input_tokens: 1366
service_tier: standard
- type: ping
- type: content_block_start
index: 0
content_block:
type: thinking
thinking: ''
- type: content_block_delta
index: 0
delta:
type: thinking_delta
thinking: The user
- type: content_block_delta
index: 0
delta:
type: thinking_delta
thinking: ' is asking me to describe what an image shows. I should describe the image objectively, focusing on'
- type: content_block_delta
index: 0
delta:
type: thinking_delta
thinking: ' visible features without identifying the person.'
- type: content_block_delta
index: 0
delta:
type: signature_delta
signature: >-
69b031ac42358c4e99f8f80f5ea48ff9cec06ca68500428ee1145e16a8020843
- type: content_block_stop
index: 0
- type: content_block_start
index: 1
content_block:
type: text
text: ''
- type: content_block_delta
index: 1
delta:
type: text_delta
text: The image shows a close-up portrait of a young girl
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >-
, probably around 3-5 years old. She has the following
features:
- **Hair
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >-
**: Wavy, light brown/auburn hair with bangs across
her forehead, with some
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
strands appearing to be pulled up
- **Eyes**: Large, expressive brown/amber eyes looking
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >2-
directly at the camera
- **Expression**: A subtle, closed-mouth smile that
gives her a gentle
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >-
, slightly mischievous look
- **Clothing**: She's wearing a white or cream-colored
top with delicate
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
lace or ruffled trim around the neckline and sleeves
- **Lighting**: The photo uses
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
warm, soft lighting that creates a gentle glow on her face
- **Background**: A muted, neutral
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
brownish-gray backdrop that keeps the focus entirely on her
The overall aesthetic has
- type: content_block_delta
index: 1
delta:
type: text_delta
text: ' a professional, portrait-photography quality with a warm, vintage feel to the color tones'
- type: content_block_delta
index: 1
delta:
type: text_delta
text: .
- type: content_block_stop
index: 1
- type: message_delta
delta:
stop_reason: end_turn
usage:
input_tokens: 1252
output_tokens: 213
cache_creation_input_tokens: 0
cache_read_input_tokens: 114
- type: message_stop
Tool Use:
value:
id: 066b133406cd8cfd88eb3b2b893a5daa
type: message
role: assistant
model: MiniMax-M3
content:
- thinking: >-
The user is asking about the weather in San Francisco.
I have access to a get_weather tool that can provide
this information. I need to format the location
parameter appropriately - the example shows "San
Francisco, US" as the format.
signature: >-
fa86efec7c27d4aa23eac60ba7a1f4e5f0f37a18d6ec4944fe495aebb6bd0930
type: thinking
- text: >-
I'll check the current weather in San Francisco for
you.
type: text
- type: tool_use
id: call_function_9pjh8jjdebck_1
name: get_weather
input:
location: San Francisco, US
usage:
input_tokens: 14
output_tokens: 91
cache_creation_input_tokens: 0
cache_read_input_tokens: 404
stop_reason: tool_use
text/event-stream:
schema:
$ref: '#/components/schemas/StreamEvent'
examples:
Stream:
value:
- type: message_start
message:
id: 066a3766d21ce23c8afc5e85a427036c
type: message
role: assistant
content: []
model: MiniMax-M3
stop_reason: null
stop_sequence: null
usage:
input_tokens: 0
output_tokens: 0
cache_creation_input_tokens: 0
cache_read_input_tokens: 1366
service_tier: standard
- type: ping
- type: content_block_start
index: 0
content_block:
type: thinking
thinking: ''
- type: content_block_delta
index: 0
delta:
type: thinking_delta
thinking: The user
- type: content_block_delta
index: 0
delta:
type: thinking_delta
thinking: ' is asking me to describe what an image shows. I should describe the image objectively, focusing on'
- type: content_block_delta
index: 0
delta:
type: thinking_delta
thinking: ' visible features without identifying the person.'
- type: content_block_delta
index: 0
delta:
type: signature_delta
signature: >-
69b031ac42358c4e99f8f80f5ea48ff9cec06ca68500428ee1145e16a8020843
- type: content_block_stop
index: 0
- type: content_block_start
index: 1
content_block:
type: text
text: ''
- type: content_block_delta
index: 1
delta:
type: text_delta
text: The image shows a close-up portrait of a young girl
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >-
, probably around 3-5 years old. She has the following
features:
- **Hair
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >-
**: Wavy, light brown/auburn hair with bangs across
her forehead, with some
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
strands appearing to be pulled up
- **Eyes**: Large, expressive brown/amber eyes looking
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >2-
directly at the camera
- **Expression**: A subtle, closed-mouth smile that
gives her a gentle
- type: content_block_delta
index: 1
delta:
type: text_delta
text: >-
, slightly mischievous look
- **Clothing**: She's wearing a white or cream-colored
top with delicate
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
lace or ruffled trim around the neckline and sleeves
- **Lighting**: The photo uses
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
warm, soft lighting that creates a gentle glow on her face
- **Background**: A muted, neutral
- type: content_block_delta
index: 1
delta:
type: text_delta
text: |2-
brownish-gray backdrop that keeps the focus entirely on her
The overall aesthetic has
- type: content_block_delta
index: 1
delta:
type: text_delta
text: ' a professional, portrait-photography quality with a warm, vintage feel to the color tones'
- type: content_block_delta
index: 1
delta:
type: text_delta
text: .
- type: content_block_stop
index: 1
- type: message_delta
delta:
stop_reason: end_turn
usage:
input_tokens: 1252
output_tokens: 213
cache_creation_input_tokens: 0
cache_read_input_tokens: 114
- type: message_stop
'400':
description: >-
Invalid request parameters (missing required fields, type not in the
whitelist, tool_use.input not a JSON object, etc.).
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: invalid_request_error
message: 'messages.0.content.1: unsupported content type ''foo'''
'401':
description: API key missing or invalid.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: authentication_error
message: API key missing or invalid.
'403':
description: No permission to access this model or path.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: permission_error
message: No permission to access this model or path.
'404':
description: Model does not exist.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: not_found_error
message: Model does not exist.
'413':
description: >-
Request body exceeds 64 MB, or a multimodal file exceeds the size
limit.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: request_too_large
message: >-
Request body exceeds 64 MB, or a multimodal file exceeds the
size limit.
'429':
description: RPM / TPM / connection-count or other rate limit triggered.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: rate_limit_error
message: RPM / TPM / connection-count or other rate limit triggered.
'500':
description: Internal server error.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: api_error
message: Internal server error.
'529':
description: Upstream model overloaded; retryable.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
example:
type: error
request_id: req_xxxxxxxx
error:
type: overloaded_error
message: Upstream model overloaded; retryable.
components:
schemas:
CreateMessageReq:
type: object
required:
- model
- messages
properties:
model:
type: string
description: >-
Model ID. MiniMax-M3 is a multimodal model with native support for
text, image, and video input, alongside tool use and thinking
content blocks. The M2.7, M2.5, M2.1, and M2 series support text and
tool calls only and do not accept image or video input.
enum:
- 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
system:
description: Set the role and behavior of the model.
oneOf:
- type: string
description: Plain text system prompt
- type: array
description: >-
System prompt in content block array format. Text blocks can
include cache_control.
items:
type: object
properties:
type:
type: string
enum:
- text
description: Content block type
text:
type: string
description: Text content
cache_control:
$ref: '#/components/schemas/CacheControl'
required:
- type
- text
messages:
type: array
description: >-
Conversation history. MiniMax-M3 supports text, image, video, tool
use, tool result, and thinking content blocks. The M2.7, M2.5, M2.1,
and M2 series support text and tool-call content blocks only; they
do not support image or video input.
items:
$ref: '#/components/schemas/Message'
stream:
type: boolean
description: >-
Whether to use streaming output, defaults to `false`. When set to
`true`, the response will be returned in chunks
default: false
max_tokens:
type: integer
format: int64
description: >-
Specifies the upper limit for generated content length (in tokens).
For 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). Content exceeding the
limit will be truncated. If generation stops due to `length`, try
increasing this value
minimum: 1
temperature:
type: number
format: double
description: >-
Temperature coefficient, affects output randomness. Range [0, 2],
default 1. Higher values produce more random output; lower values
produce more deterministic output.
minimum: 0
maximum: 2
default: 1
top_p:
type: number
format: double
description: >-
Nucleus sampling parameter. Range [0, 1]. Default is 0.95 for
MiniMax-M3 and 0.9 for M2.x models.
minimum: 0
maximum: 1
default: 0.95
tools:
type: array
description: Tool definitions for Anthropic-compatible tool use.
items:
$ref: '#/components/schemas/Tool'
tool_choice:
$ref: '#/components/schemas/ToolChoice'
thinking:
type: object
description: >-
Controls MiniMax-M3 thinking. When omitted, thinking is disabled by
default and responses do not include thinking blocks. For M2.x
models, thinking cannot be disabled.
properties:
type:
type: string
enum:
- disabled
- adaptive
default: disabled
description: >-
Thinking control type.
- `disabled`: Disable MiniMax-M3 thinking output. This is the
default when `thinking` is omitted. For M2.x models, thinking
remains on.
- `adaptive`: Enable MiniMax-M3 thinking output and return
thinking blocks.
required: []
metadata:
type: object
description: >-
Request metadata. user_id is recommended for end-user-level
aggregation, rate limiting, and billing analysis.
properties:
user_id:
type: string
description: End-user ID.
CreateMessageResp:
type: object
properties:
id:
type: string
description: Unique ID of this response
type:
type: string
description: Object type, fixed as `message`
enum:
- message
role:
type: string
description: Role, fixed as `assistant`
enum:
- assistant
model:
type: string
description: Model ID used for this request
content:
type: array
description: List of response content blocks
items:
$ref: '#/components/schemas/ResponseContentBlock'
stop_reason:
type: string
description: |-
Reason for stopping generation:
- end_turn: Model ended naturally
- max_tokens: Reached max_tokens limit
- tool_use: Model requested tool use
enum:
- end_turn
- max_tokens
- tool_use
usage:
$ref: '#/components/schemas/Usage'
StreamEvent:
type: object
description: ''
required:
- type
properties:
type:
type: string
description: >-
Event type:
- `message_start`: Message start, contains complete message metadata
- `ping`: Heartbeat event
- `content_block_start`: Content block start
- `content_block_delta`: Content block incremental update
- `content_block_stop`: Content block end
- `message_delta`: Message-level incremental update (e.g.,
stop_reason)
- `message_stop`: Message end
enum:
- message_start
- ping
- content_block_start
- content_block_delta
- content_block_stop
- message_delta
- message_stop
message:
type: object
description: Message object (returned when `type` is `message_start`)
properties:
id:
type: string
description: Unique ID of the message
type:
type: string
enum:
- message
role:
type: string
enum:
- assistant
content:
type: array
description: Content block list, initially an empty array
items:
$ref: '#/components/schemas/ResponseContentBlock'
model:
type: string
description: Model ID
stop_reason:
type: string
nullable: true
description: Stop reason, null at the start of streaming
stop_sequence:
type: string
nullable: true
description: Stop sequence, null at the start of streaming
usage:
$ref: '#/components/schemas/Usage'
index:
type: integer
description: >-
Index of the content block (returned for `content_block_start`,
`content_block_delta`, `content_block_stop`)
content_block:
$ref: '#/components/schemas/ResponseContentBlock'
description: Content block object (returned when `type` is `content_block_start`)
delta:
type: object
description: >-
Incremental update content (returned for `content_block_delta` or
`message_delta`)
properties:
type:
type: string
description: >-
Delta type, for example text_delta, thinking_delta, or
signature_delta.
enum:
- text_delta
- thinking_delta
- signature_delta
text:
type: string
description: Incremental text content
stop_reason:
type: string
description: Reason for stopping generation.
enum:
- end_turn
- max_tokens
- tool_use
usage:
$ref: '#/components/schemas/Usage'
description: Token usage (returned for `message_delta`)
ErrorResponse:
type: object
description: >-
Error response. Uses an HTTP status code plus a JSON body. When an error
occurs during streaming, it is delivered as an `event: error` SSE event
with the same body structure; clients should stop reading and clean up
the session state after receiving an error.
properties:
type:
type: string
enum:
- error
description: Always `error`.
request_id:
type: string
description: Unique identifier of the request, useful for troubleshooting.
error:
type: object
properties:
type:
type: string
description: Error type.
enum:
- invalid_request_error
- authentication_error
- permission_error
- not_found_error
- request_too_large
- rate_limit_error
- api_error
- overloaded_error
message:
type: string
description: Error details.
CacheControl:
type: object
description: Prompt cache marker.
properties:
type:
type: string
enum:
- ephemeral
description: Prompt cache marker.
required:
- type
Message:
type: object
required:
- role
- content
properties:
role:
type: string
enum:
- user
- assistant
- user_system
- group
- sample_message_user
- sample_message_ai
description: >-
Role of the message sender. MiniMax-M3 uses alternating user /
assistant messages.
content:
description: >-
Message content. MiniMax-M3 supports text, image, video, tool use,
tool result, and thinking content blocks.
oneOf:
- type: string
description: Plain text message
- type: array
description: >-
Message content. MiniMax-M3 supports text, image, video, tool
use, tool result, and thinking content blocks.
items:
$ref: '#/components/schemas/RequestContentBlock'
Tool:
type: object
description: Tool definitions for Anthropic-compatible tool use.
properties:
name:
type: string
description: Tool name.
description:
type: string
description: Tool description.
input_schema:
type: object
description: Tool input parameters JSON Schema.
cache_control:
$ref: '#/components/schemas/CacheControl'
required:
- name
- input_schema
ToolChoice:
type: object
description: Tool selection strategy. Only auto and none are supported.
properties:
type:
type: string
enum:
- auto
- none
description: Tool choice type.
required:
- type
ResponseContentBlock:
type: object
description: >-
Content block returned in response messages. The model emits text,
tool_use (when calling tools), and thinking (when extended thinking is
active). Responses never contain image, video, tool_result, or
mid_conv_system blocks.
required:
- type
properties:
type:
type: string
description: |-
Content block type:
- text: Text content
- tool_use: Model tool call
- thinking: Model thinking content
enum:
- text
- tool_use
- thinking
text:
type: string
description: Text content when type=text.
id:
type: string
description: Tool call ID returned when type=tool_use.
name:
type: string
description: Tool name returned when type=tool_use.
input:
type: object
description: Tool input returned when type=tool_use.
thinking:
type: string
description: Model thinking content when type=thinking.
signature:
type: string
description: Signature for thinking content. Return unchanged in later turns.
Usage:
type: object
description: >-
Token usage for this request, including prompt cache usage when
applicable.
properties:
input_tokens:
type: integer
description: Number of tokens consumed by input
output_tokens:
type: integer
description: Number of tokens consumed by output
cache_creation_input_tokens:
type: integer
description: Number of input tokens used to create prompt cache entries.
cache_read_input_tokens:
type: integer
description: Number of input tokens read from prompt cache.
RequestContentBlock:
type: object
description: >-
Content block accepted in request messages. text and tool blocks are
supported by all models; image and video blocks are MiniMax-M3 only.
thinking is only used when echoing a prior assistant turn back to the
model.
required:
- type
properties:
type:
type: string
description: |-
Content block type:
- text: Text content
- image: Image input, MiniMax-M3 only
- video: Video input, MiniMax-M3 only
- tool_use: Echoed assistant tool call from a previous turn
- tool_result: Tool execution result
- thinking: Echoed assistant thinking from a previous turn
- mid_conv_system: System instructions inserted mid-conversation
enum:
- text
- image
- video
- tool_use
- tool_result
- thinking
- mid_conv_system
text:
type: string
description: Text content when type=text.
source:
$ref: '#/components/schemas/MediaSource'
description: >-
Image or video source when type=image or type=video. MiniMax-M3
only.
id:
type: string
description: Tool call ID when echoing type=tool_use.
name:
type: string
description: Tool name when echoing type=tool_use.
input:
type: object
description: Tool input when echoing type=tool_use.
tool_use_id:
type: string
description: ID of the matching tool_use block when type=tool_result.
content:
oneOf:
- type: string
- type: array
items:
$ref: '#/components/schemas/RequestContentBlock'
description: >-
Tool execution result, either a string or an array of text/image
content blocks.
thinking:
type: string
description: Echoed assistant thinking content when type=thinking.
signature:
type: string
description: >-
Signature for echoed thinking content. Return unchanged in later
turns.
cache_control:
$ref: '#/components/schemas/CacheControl'
MediaSource:
type: object
description: >-
Media source for image or video blocks.
**Supported image formats**
| Format | Common extensions | MIME Type |
| :-- | :-- | :-- |
| JPEG | .jpg, .jpeg | image/jpeg |
| PNG | .png | image/png |
| GIF | .gif | image/gif |
| WEBP | .webp | image/webp |
**Supported video formats**
| Format | Common extensions | MIME Type |
| :-- | :-- | :-- |
| MP4 | .mp4 | video/mp4 |
| AVI | .avi | video/avi or video/x-msvideo |
| MOV | .mov | URL input: set the object storage Content-Type to
video/quicktime; base64 input: use video/mov, i.e.
data:video/mov;base64, |
| MKV | .mkv | video/x-matroska |
**File size and request limits**
- URL or Base64 input: video file ≤ 50 MB, image file ≤ 10 MB, request
body ≤ 64 MB.
- Files API upload: referenced as `mm_file://{file_id}`, with a single
video up to 512 MB.
properties:
type:
type: string
enum:
- base64
- url
description: Media source type.
media_type:
type: string
description: Required for base64 input, for example image/png or video/mp4.
data:
type: string
description: Base64-encoded media bytes.
url:
type: string
description: >-
Public URL. Videos can also use a Files API reference in the form
`mm_file://{file_id}`. To obtain a `file_id`, first upload the video
via the [Upload File](/api-reference/file-management-upload) API.
detail:
type: string
enum:
- low
- default
- high
default: default
description: >-
Understanding detail level. Defaults to default.
For image input, rough single-image token usage heuristic:
| detail | Rough token usage |
| :-- | :-- |
| low | Usually a few hundred tokens, up to ~600 |
| default | Often ~1k-3k tokens, up to ~5k |
| high | Often several thousand tokens, up to ~15k+ |
Actual usage depends on image size and content; use `POST
/anthropic/v1/messages/count_tokens` or response usage for exact
usage.
fps:
type: number
minimum: 0.2
maximum: 5
default: 1
description: >-
Video frame sampling rate. Defaults to 1. Range [0.2, 5].
- Higher values: more sensitive to scene changes, but higher token
cost and slower response.
- Lower values: lower token cost and faster response, but less
sensitive to scene changes.
max_long_side_pixel:
type: integer
minimum: 1
description: Longest-side pixel limit for the image or video frames.
required:
- type
securitySchemes:
bearerAuth:
type: http
scheme: bearer
description: >-
Bearer API Key auth. Send Authorization: Bearer . If
Authorization and x-api-key are both present, Authorization takes
precedence.
apiKeyAuth:
type: apiKey
in: header
name: x-api-key
description: >-
Anthropic-compatible API Key auth. Send x-api-key: .
Authorization: Bearer is recommended.
````