MiniMax M3

Generic Info

• Publisher: MiniMax AI
• Release Date: 2026
• Architecture: Omnimodal Mixture-of-Experts (MoE) with Adaptive Thinking
• Context Window: 1,000,000 tokens (1M context)
• License: MiniMax Community Open License
• Key Capabilities: Video Understanding, Multimodal Agent Workflows, SWE-bench Verified Coding, Native Function Calling

MiniMax M3 is MiniMax's landmark open-weights release, delivering unified omnimodal processing across text, images, audio, and extended video files. Built with sparse MoE routing and an adaptive reasoning engine, M3 achieves top scores on SWE-bench Verified and competitive coding leaderboards while retaining high-efficiency inference. Its native XML tool calling schema allows reliable invocation of external APIs and robotic tools.

Hello World Guide

Run MiniMax M3 locally using Hugging Face transformers.

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "MiniMaxAI/MiniMax-M3"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)

prompt = "Analyze this system log snippet and write an automated diagnostic script."
messages = [
    {"role": "system", "content": "You are MiniMax-M3, an omnimodal reasoning and coding model."},
    {"role": "user", "content": prompt}
]

inputs = tokenizer.apply_chat_template(
    messages,
    return_tensors="pt",
    add_generation_prompt=True
).to(model.device)

outputs = model.generate(
    inputs,
    max_new_tokens=512,
    temperature=0.6
)

response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)
print(response)

Industry Usage