Everything about MiniMax-M3

Overview

Released on June 1, 2026, MiniMax-M3 is one of the 6 MiniMax models tracked on this site. It is an open-weight model: you can host it yourself on infrastructure you control, and 52 API providers listed on models.dev also offer it. It has a reasoning mode: it can "think" before answering, which helps with multi-step problems but uses more tokens, and therefore more money. That makes it 4 months old as of October 22, 2026.

We file it under open-weight models (68 models in this category on the site).

Pricing and budget

MiniMax-M3 costs $0.30 per million input tokens and $1.2 per million output tokens (median of 52 providers, checked on October 22, 2026).

In practice, an agent task that reads 20,000 tokens and writes 2,000 costs about $0.008, or $8.4 for 1,000 such tasks.

That is 1.2 times the median output price of open-weight models ($1).

To estimate your own budget, use the agent cost calculator.

What do independent benchmarks say?

According to results published by Epoch AI (checked on October 22, 2026), MiniMax-M3 has a capability index of 55.5 out of 100, computed on 4 benchmarks from our basket.

It is not among the top 30 of the weekly ranking.

• GPQA Diamond (PhD-level science questions): 90.9% — the best tracked model, Gemini 4 Argon, reaches 96.4%
• OTIS Mock AIME (competition mathematics problems): 71.1% — the best tracked model, GPT-5.5 Pro, reaches 100%
• FrontierCode (advanced programming tasks): 14.7% — the best tracked model, Claude Opus 5.5, reaches 54.6%
• APEX-Agents (professional tasks carried out by an agent): 37.7% — the best tracked model, Gemini 4 Argon, reaches 82.4%

Technical specifications

Its context window reaches 1,048,576 tokens, roughly 1,573 pages of text at 500 words per page: enough to load a whole book or a mid-sized codebase. A single answer can run to 50,000 tokens.

It accepts as input: text, image and video.

It can call tools (functions, APIs), which makes it a possible foundation for an AI agent.

The weights are published on Hugging Face (MiniMaxAI/Minimax-M3). See also our list of open-source models.

Who is it for?

Given its category, MiniMax-M3 is mainly aimed at projects where you want control over your data or hosting, or want to adapt the model to your field.

Its very large context window makes it worth considering for analysing large sets of documents in a single request.

Before adopting it, test it on your own examples: benchmarks measure standard tasks, not your job.

What could replace it?

In the same category, the models closest to MiniMax-M3 are:

• Qwen3.6 35B-A3B (Alibaba Cloud (Qwen), index 55.3)
• Qwen3.5 397B-A17B (Alibaba Cloud (Qwen), index 59.2)
• DeepSeek V3.2 (DeepSeek, index 59.8)
• Kimi K2.7 Code (Moonshot AI, index 66.7)