MiniMax M3
By MiniMax · China
Updated 2026-08-28
Overview
MiniMax's frontier multimodal MoE, 427B total parameters, combining vision, code, and reasoning with context up to 1M tokens. Firmly datacenter-scale hardware territory.
When to pick this model
- Enterprise deployments with multi-GPU server capacity for frontier-scale multimodal reasoning
- Vision-plus-text workflows combined with code generation
- Long-document or long-context analysis up to 1M tokens
- Teams already running via Ollama's official tag
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 248 GB |
| Q5_K_M | 303 GB |
| Q8_0 | 457 GB |
| FP16 (no quantization) | 854 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, MiniMax M3 is server-class even at Q4_K_M (248 GB). Stepping up to Q8_0 nearly doubles the footprint to 457 GB, and unquantized FP16 weights take 854 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, MiniMax M3 needs roughly 555 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 1.5 tokens/sec on entry-level GPUs, on the order of 2.5 tokens/sec on a mid-range card, and up to 5 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches MiniMax M3 to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for MiniMax M3 |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | Does not fit — needs 248 GB at Q4_K_M |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | Does not fit — needs 248 GB at Q4_K_M |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | Does not fit — needs 248 GB at Q4_K_M |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | Does not fit — needs 248 GB at Q4_K_M |
| 32 GB | RTX 5090 | Does not fit — needs 248 GB at Q4_K_M |
Which GPU should you buy to run MiniMax M3?
To run MiniMax M3 locally at Q4, you need ~248 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Strengths
- Multimodal: vision plus text in one model
- Strong code and reasoning performance
- Massive context window up to 1M tokens
- Available via an official Ollama tag
Limitations
- Very heavy: ~248GB VRAM at Q4, multi-GPU servers only
- Low local throughput (~5 tok/s at Q4)
- Custom "other" license — review usage terms before deploying
Typical workloads
In our catalog grid, MiniMax M3 is filed under Multimodal Reasoning, Code Generation, Long Documents — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: code generation and review (pair it with an editor integration like Continue.dev or Cline); multi-step reasoning and math-flavoured tasks; vision-language work — screenshots, charts, scanned documents; multilingual workloads.
The 1024k-token context window is large enough to hold entire codebases' worth of files or long reports in a single prompt, which is what makes local RAG and document analysis practical. It ships under the MiniMax (custom) license — commercial use is generally possible but read the specific terms before embedding it in a product.
Architecture & training
Architecture: Multimodal MoE · 427B total parameters · context window up to 1M tokens
Training: Multimodal MoE model from MiniMax, pretrained for vision, code, and long-context reasoning. Training details not published.
A frontier-scale multimodal MoE for enterprises with the GPU budget to match — not for local or single-GPU use.
Quick start
ollama run minimax-m3Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does MiniMax M3 need?
At the recommended Q4_K_M quantization, MiniMax M3 needs about 248 GB of VRAM. Q8_0 takes 457 GB, and unquantized FP16 weights take 854 GB.
Can MiniMax M3 run without a GPU?
Yes — with roughly 555 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does MiniMax M3 support?
MiniMax M3 supports a 1024k-token context window (1,048,576 tokens).
Can I use MiniMax M3 commercially?
MiniMax M3 ships under the MiniMax (custom) license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.
How fast is MiniMax M3 on consumer hardware?
Our compatibility engine estimates on the order of 2.5 tokens/sec on a mid-range GPU and up to 5 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of MiniMax M3 should I download first?
Start with Q4_K_M (248 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.