MiniMax
Operational· checked every few minutes

MiniMax M3 API

The MiniMax M3 API for coding agents and long context, 57% below MiniMax's list price

Price per 1M tokens57%off
Input
$0.13$0.30
Output
$0.51$1.20
Cached input
$0.026$0.06
Model IDminimax-m3Updated
chat.py
from openai import OpenAI
client = OpenAI(
base_url="https://api.zurelay.com/v1",
api_key="YOUR_ZURELAY_KEY",
)
stream = client.chat.completions.create(
model="minimax-m3",
messages=[
{"role": "user", "content": "Explain quicksort briefly."},
],
stream=True,
)
for chunk in stream:
if chunk.choices:
print(chunk.choices[0].delta.content or "", end="")

Base URL https://api.zurelay.com/v1. Works with any OpenAI SDK. Chat API docs

Context window
1M tokens
Reasoning
none · high
Input
Text, Image
Released
Jun 1, 2026
Status
Operational

Pricing

MiniMax M3 API pricing. A fraction of the list price.

Pay as you go from prepaid credit, with no subscription. Top up from $10. Requests that fail are never billed.

RateZurelayMiniMax list
Input
per 1M tokens
$0.13$0.30
Output
per 1M tokens
$0.51$1.20
Cached input
per 1M tokens
$0.026$0.06

One rate at every prompt length, up to the full 1M context window. Streaming, tool calls and structured outputs cost nothing extra.

Every MiniMax model and price
Savings calculator

Move the sliders to your monthly usage.

Input tokens per month50M
Output tokens per month10M
Zurelay
$11.60
MiniMax list price
$27.00
You save every month$15.40

$184.80 a year

Cost

MiniMax M3 API cost, worked out.

Typical workloads at Zurelay's live prices, next to the list price.

  • Chatbot

    50,000 requests a month, 1,500 tokens in and 300 out, 30% of the input cached. $0.000301 a request.

    $15.06a month$35.10 at list
  • Documents

    20,000 requests a month, 8,000 tokens in and 500 out, 20% of the input cached. $0.00113 a request.

    $22.57a month$52.32 at list
  • Coding agent

    5,000 requests a month, 40,000 tokens in and 1,500 out, 85% of the input cached. $0.00243 a request.

    $12.15a month$28.20 at list
  • Classification

    1,000,000 requests a month, 400 tokens in and 5 out. $0.000055 a request.

    $54.55a month$126.00 at list

Overview

What is MiniMax M3?

MiniMax M3 is MiniMax's natively multimodal model for agentic coding, tool use and long context, released on June 1, 2026, with about 428B total and 23B active parameters. The Zurelay MiniMax M3 API serves it through one OpenAI-compatible endpoint with a 1M-token context window and image input, at 57% below MiniMax's list price.

MiniMax released M3 on June 1, 2026 as the MiniMax-M3 API model, and has since published the weights on Hugging Face. It is a Mixture-of-Experts model with about 428B total parameters, of which about 23B are active per token. MiniMax calls it the first open-weight model to combine frontier coding, a 1M-token context and native multimodality. On Zurelay the MiniMax M3 API model ID is minimax-m3, and it costs $0.13 per 1M input tokens and $0.51 per 1M output tokens.

MiniMax's reported results center on coding and long agent runs. M3 scores 59.0% on SWE-Bench Pro, 80.5% on SWE-Bench Verified and 66.0% on Terminal Bench 2.1, against 56.2%, 79.9% and 51.1% for M2.7. It reaches 83.5% on BrowseComp and 74.2% on MCP Atlas. In MiniMax's own long-horizon tests, M3 worked for nearly 12 hours to reproduce the core experiments of an ICLR 2025 paper, and spent about 24 hours and 1,959 tool calls making an FP8 GEMM kernel 9.4 times faster.

M3 replaces full attention with MiniMax Sparse Attention (MSA), which scores blocks of the context and attends only to the most relevant ones. MiniMax reports 9 times faster prefill and 15 times faster decoding than M2 at a 1M-token context, with per-token compute cut to a twentieth. M3 was trained on text, images and video together from the first step of pretraining. On Zurelay it reads text and images and writes text, with a context window of 1,048,576 tokens.

M3 is a reasoning model that can think between tool calls, so in multi-turn agent loops MiniMax asks you to send each full assistant message back. MiniMax recommends temperature 1.0 and top_p 0.95. It charges a higher rate when a request has more than 512K input tokens; Zurelay charges one rate at every prompt length, up to the full context window. The weights come under the MiniMax Community License, which permits commercial use with conditions.

Strengths

Where MiniMax M3 shines. And what teams build with it.

01

Frontier coding scores

MiniMax reports 59.0% on SWE-Bench Pro, 80.5% on SWE-Bench Verified and 66.0% on Terminal Bench 2.1, ahead of M2.7 on all three.

02

Long context that stays fast

MiniMax Sparse Attention makes prefill 9 times faster and decoding 15 times faster than M2 at a 1M-token context.

03

Reads images natively

Trained on images from the first step of pretraining. MiniMax reports 78.1 on MMMU-Pro and 91.6 on OmniDocBench, a document parsing benchmark.

04

Built for long agent runs

In MiniMax's tests, M3 worked for nearly 12 hours to reproduce a research paper, and made 1,959 tool calls over about 24 hours to tune a GPU kernel.

Use cases

  • Coding agents

    Multi-file fixes, refactors and terminal tasks in agent harnesses, with reasoning carried between tool calls.

  • Whole-repository questions

    Load a large codebase, long logs or a full agent trace into the 1M-token window and ask across all of it.

  • Screenshots and documents

    Read UI screenshots, charts and scanned pages next to code or text, in the same request.

  • Research agents

    Browse, search and call tools over many steps. MiniMax reports 83.5% on BrowseComp.

Get started

Call MiniMax M3 in three steps.

No waitlist and no new SDK. If your code already talks to OpenAI, it already talks to Zurelay.

  1. 1

    Create a key

    Sign up, add credit and create an API key. Set a monthly budget or a rate limit per key if you like.

  2. 2

    Point your SDK at Zurelay

    Change the base URL. Everything else in your code stays the same.

    https://api.zurelay.com/v1
  3. 3

    Use minimax-m3

    Send chat completions as usual. Streaming, tool calls and usage reporting work as you expect.

    minimax-m3

FAQ

MiniMax M3 API questions.

Anything else? Ask our team and we’ll answer by email.

How much does the MiniMax M3 API cost?

On Zurelay, minimax-m3 costs $0.13 per 1M input tokens and $0.51 per 1M output tokens, with cached input at $0.026. MiniMax's list price is $0.30 input and $1.20 output per 1M, so you save 57%. Reasoning tokens count as output.

Is there a cheaper MiniMax M3 API than MiniMax's own?

Yes. Zurelay serves minimax-m3 at 57% below MiniMax's list price, and one rate covers every prompt length, while MiniMax charges more for requests over 512K input tokens. You pay from prepaid credit with no subscription, and failed requests are never billed.

How do I call the MiniMax M3 API with the OpenAI SDK?

Install the official openai package, set base_url to https://api.zurelay.com/v1 and use your Zurelay API key. Then send a Chat Completions request with model set to "minimax-m3". Zurelay also accepts MiniMax's own name, MiniMax-M3, and minimax/minimax-m3, so existing code only needs a new base URL and key.

What is the MiniMax M3 context window?

1,048,576 tokens, which MiniMax describes as a 1M-token context. On Zurelay the price per token stays the same up to the full window. M3 takes text and images as input and writes text.

Can MiniMax M3 read images?

Yes. Add image_url parts with an https link or a base64 data URL; PNG, JPEG, WebP and GIF work, and one message can hold several images. Images count as input tokens, and MiniMax's guide puts a typical image at about 1,000 to 3,000 tokens. MiniMax's model can also read video, but on Zurelay M3 takes text and images only; for video input, use Gemini 3.7 Flash.

Is MiniMax M3 open source?

The weights are public on Hugging Face under the MiniMax Community License rather than a standard open-source license. Non-commercial use is free. Commercial use requires a visible "Built with MiniMax M3" credit and a one-time notice to MiniMax, and products earning more than 20 million US dollars a year need MiniMax's written authorization first.

What is MiniMax M3 best at?

Agentic coding, tool calling and long-context work: multi-file changes and terminal tasks in agent harnesses, research agents that browse and call tools over many steps, and questions across a whole repository or document set. It also reads screenshots, charts and scanned pages.

How is MiniMax M3 different from MiniMax M2.7?

M3 reads images as well as text, has a 1,048,576-token context window against 204,800 for M2.7, and uses MiniMax Sparse Attention to keep long prompts fast. MiniMax's comparison table puts M3 ahead of M2.7 on every benchmark it lists for both, for example 59.0% against 56.2% on SWE-Bench Pro and 74.2% against 49.4% on MCP Atlas. Zurelay no longer serves M2.7, and MiniMax retires it on November 18, 2026.

Does the MiniMax M3 API support tool calling and streaming?

Yes. Pass tools as you would to any OpenAI model, and send each full assistant message back in the conversation so M3 keeps its reasoning between calls, as MiniMax recommends. Set stream: true, and add stream_options.include_usage to get token counts in the final chunk. Each Zurelay API key can have its own monthly budget and requests-per-minute cap.

Is this the real MiniMax M3?

Yes. Requests for minimax-m3 run on MiniMax's M3, and Zurelay doesn't quietly swap in a different model: if you turn on smart routing and M3 is down, the response names the model that answered. Responses are sampled, so wording varies between runs. Zurelay is independent and is not affiliated with or endorsed by MiniMax.

More models on Zurelay

All models

Stop paying list price.
Start saving today.

Create a key in seconds, point the SDK you already use at Zurelay, and every request costs up to 90% less from the first token.