MiniMax-M3

The first open-weight model combining frontier agentic coding, a million-token context and native multimodality; MSA sparse attention cuts long-context inference cost to about 1/20 of the previous generation.

Specifications

Item	Value
Model ID	`MiniMax-M3`
Vendor	MiniMax
Vendor release	2026-06-01
Available on APIYI	2026-06-05
Knowledge cutoff	Not disclosed
Input modalities	Text, Image, Video
Output modalities	Text
Context window	1,000,000 tokens
Billing	Usage-based

Pricing

Prices in USD per 1M tokens ($/1M).

Input	Cached input	Output
$0.3	$0.06	$1.2

The table shows list prices. Recharge promotions and group discounts stack; the console reflects the actual charge in real time. See Pricing and Recharge promotions.

Tiered pricing

This model is tier-priced by the token size of each request (output price = tier input price × output multiplier):

0 – 524,288 tokens: $0.3/1M input
Above 524,288 tokens: $0.6/1M input

(already discounted to 50% of the vendor’s list price)

Endpoints

Endpoint	Path	Supported
`OpenAI Chat Completions`	`POST /v1/chat/completions`	✅
`OpenAI Responses`	`POST /v1/responses`	—
`Anthropic Messages`	`POST /v1/messages`	✅
`Gemini Generate Content`	`POST /v1beta/models/{model}:generateContent`	—
`Image Generations`	`POST /v1/images/generations`	—
`Embeddings`	`POST /v1/embeddings`	—

Billing groups

Group	Multiplier	Notes
`ClaudeCode`	0.95×	95% of list price
`Default`	1×	List price
`SVIP`	1×	List price

Some groups carry extra discounts, stackable with recharge bonuses. See Tokens and groups.

Supported features

Feature	Supported
Streaming	✅
Tool calling	✅
Structured outputs	✅
Vision	✅
Prompt caching	✅
Extended thinking	Opt-in

Example request

The example below calls MiniMax-M3 through OpenAI Chat Completions (/v1/chat/completions). Point base_url at https://api.apiyi.com/v1 — everything else matches the official API.

curl https://api.apiyi.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $APIYI_API_KEY" \
  -d '{
    "model": "MiniMax-M3",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Read the key from an environment variable, never hard-coded. In production, issue separate tokens per use case so you can revoke and attribute usage individually.

Launch announcement

Background, benchmarks and migration notes for MiniMax-M3

Model pricing directory

Live pricing, endpoints and groups for all 308 models

Specs on this page are maintained by hand in models/data/model-details.json; pricing and endpoints come from the live pricing API, updated 2026-07-31 01:30 (UTC+8), specs last verified 2026-07-31.

Pricing Table

OpenAI

Anthropic

Google

xAI

Alibaba

ByteDance

Moonshot

Zhipu

MiniMax

Specifications

Pricing

Tiered pricing

Endpoints

Billing groups

Supported features

Example request

Launch announcement

Model pricing directory

​Specifications

​Pricing

​Tiered pricing

​Endpoints

​Billing groups

​Supported features

​Example request

​Related docs

Launch announcement

Model pricing directory

Specifications

Pricing

Tiered pricing

Endpoints

Billing groups

Supported features

Example request

Related docs