minimax-m3 API — per-token pricing and code examples

Checking statusPrice version 4 · in effect since Nov 12, 2026

Price per 1M tokens

InputOfficial price $0.30$0.03
Cache readOfficial price $0.06$0.0072
Cache write$0.03
OutputOfficial price $1.20$0.12

Minimum per request$0.00002

Cache rates apply when the provider reports cache tokens.

90% off

About this model

minimax-m3 is served through OpenAI Chat Completions, OpenAI Responses, Anthropic Messages. You pay $0.03 per 1M input tokens and $0.12 per 1M output tokens. That is 90% below the official price of $0.30 and $1.20. Cache reads cost $0.0072 per 1M tokens when the provider reports cached tokens. It is included in budget packs from $1.99 up.

Where MiniMax M3 fits

MiniMax describes M3 as a frontier multimodal coding model and sums it up as "Coding & Agentic Frontier. 1M-context MSA. Native Multimodality." The maker presents minimax-m3 as a reliable foundation for AI coding assistants and automated workflows, and also points to long-horizon agent work and understanding long videos. MiniMax says the M3 API supports up to 1M tokens of context, with a guaranteed minimum of 512K tokens. Input can be text, images in JPEG, PNG, GIF or WEBP, and video; audio input is not supported yet, and output is text. MiniMax does not publish a separate output ceiling for M3, and it notes that thinking tokens are counted inside max_tokens or max_completion_tokens, so a tight limit can be consumed before the visible answer appears.

Optional thinking on MiniMax M3

Unlike the newer Flash model, MiniMax M3 lets you choose whether it thinks before answering. The reasoning_effort field does not apply to M3, so depth is not tuned that way; the choice is on or off. Because MiniMax's OpenAI-format and Anthropic-format references describe this switch differently, the reliable approach is to state what you want in every request instead of relying on the default. In the OpenAI format, MiniMax documents thinking: {"type": "disabled"} as the way to skip thinking and answer directly. In the Anthropic format, MiniMax enables thinking on M3 with the adaptive type. MiniMax also documents that in the OpenAI format the thinking is returned inside content as a <think>…</think> tag, unless the request sends reasoning_split: true, which moves it to a separate reasoning_content field.

Tool calling works with interleaved thinking, meaning minimax-m3 can reason between tool calls. MiniMax asks that the complete assistant message be returned on the next turn so that chain is not broken. A few parameters are worth knowing about: temperature and top_p are not adjustable yet and stay at 1 and 0.95, n only accepts 1, the legacy function_call field is not supported, and on the Anthropic format top_k and stop_sequences are ignored.

What it costs on Cent AI

MiniMax M3 is billed per token from your prepaid USD balance using the input, cache read, cache write and output rates in the price box on this page. When thinking is on, Cent AI bills those reasoning tokens as output, so switching thinking off for simple extraction or formatting tasks is the quickest way to reduce output spend on this model.

MiniMax caches prompts automatically for M3, with no configuration needed. Caching applies to calls with 512 or more input tokens, so very short prompts are never cached. MiniMax does not publish a cache lifetime for M3. Cent AI charges cache rates only for cached tokens that MiniMax reports in the usage of the request.

Using it from your tools

minimax-m3 is available on /v1/chat/completions, /v1/responses and /v1/messages, and Cent AI does not change your parameters silently. If an old integration sets temperature: 0.3 and expects it to matter, it will not on M3. The example below turns thinking off explicitly, in the /v1/chat/completions format:

{
  "model": "minimax-m3",
  "thinking": { "type": "disabled" },
  "max_completion_tokens": 2048,
  "messages": [{ "role": "user", "content": "Convert this table to CSV." }]
}

On /v1/messages, use the thinking block with the adaptive type when you want MiniMax M3 to reason, and leave room in max_tokens for the thinking it produces.

Specification

Context window
—
Input and output
Text
Endpoints
/v1/chat/completions, /v1/responses, /v1/messages
Smallest pack
$1.99

Latency and error rate per model are on the status page. status.cent-ai.dev

What a request costs

Open the full calculator

Cost per request

90% is an assumption you can change, not a promise. Cache is billed only from cache tokens the model provider reports.

See how many such requests it covers.

Estimate

minimax-m3
Fresh input99,000 tok
Cache read891,000 tok
Output10,000 tok

Per request$0.0105
At official pricesAt official prices $0.0951

Total$0.0105

An estimate at price version 4. Your bill comes from the usage the provider reports for each request.

Save $0.0845

Code

curl https://api.cent-ai.dev/v1/chat/completions \
  -H "Authorization: Bearer $CENT_AI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "minimax-m3", "messages": [{"role": "user", "content": "Hello"}]}'

Set CENT_AI_KEY to a key from the dashboard. OpenAI SDKs use https://api.cent-ai.dev/v1; Anthropic SDKs use https://api.cent-ai.dev.

Price history

Version 4, in effect since Nov 12, 2026.

  1. Effective

    per 1M tokens

    Inputold $0.03 new $0.03
    Cache readold $0.003 new $0.0072
    Cache writeold $0.03 new $0.03
    Outputold $0.12 new $0.12
  2. Effective

    per 1M tokens

    Inputold $0.015 new $0.03
    Cache readold $0.0015 new $0.003
    Cache writeold $0.015 new $0.03
    Outputold $0.06 new $0.12
  3. Effective

    per 1M tokens

    Inputold $0.015 new $0.015
    Cache readold $0.003 new $0.0015
    Cache writeold $0.015 new $0.015
    Outputold $0.06 new $0.06

Compare with

All models against official prices

Questions

How much does minimax-m3 cost on Cent AI?

$0.03 per 1M input tokens and $0.12 per 1M output tokens. Cache reads cost $0.0072 and cache writes $0.03 per 1M tokens. Each request is rounded up once.

Is there a minimum charge?

A successful request costs at least $0.00002. A request that ends early, because the client left or the output was cut off, pays only for its tokens.

How do I pay?

Top up a prepaid USD balance with QRIS or USDT/USDC, or buy a budget pack once. Nothing renews automatically.

What if the provider does not report usage?

The request is billed from an estimate and marked estimated on the Requests page. An estimate never contains cache tokens.

Which endpoint do I call?

/v1/chat/completions, /v1/responses, /v1/messages. The OpenAI base URL is https://api.cent-ai.dev/v1.

Try minimax-m3 with one key

One prepaid balance for every model. Pay with QRIS or USDT/USDC.

Start free