MiniMax M3 VRAM Requirements

MiniMax M3 at a glance

Made by

MiniMax, China

Released

June 2, 2026

Parameters

427.0B, 23B active

Context window

1M tokens

License

Not declared in its Hugging Face metadata: see the model card (opens in a new tab)

Reads

Text, images and video

Tool calling

Supported

Reasoning

Has a reasoning mode

On ModelScope

MiniMax/MiniMax-M3 (opens in a new tab)

On Ollama

ollama run minimax-m3 0 pulls (opens in a new tab)

VRAM by precision and context length

Weights plus the KV cache of one request (BF16 cache), in GiB. Add 1 to 3 GiB for the inference engine itself.

How many GPUs to run MiniMax M3

Fewest GPUs that fit the weights plus one 8K-token request. Data-center GPUs use vLLM (90% of memory usable), the others llama.cpp. Several GPUs split the model with tensor or pipeline parallelism.

8-bit: FP8 with vLLM (INT8 on GPUs without FP8), Q8_0 with llama.cpp. 4-bit: AWQ / GPTQ with vLLM, Q4_K_M with llama.cpp, both counted at the Q4_K_M size. Not on this GPU: the official MXFP8 weights do not run there (llama.cpp needs a GGUF version, and vLLM needs a GPU with FP8 support).

Cheapest way to run MiniMax M3 today

For each GPU: the fewest cards that fit MiniMax M3 with one 8K-token request, the most faithful weight format at that count, and the cheapest on-demand price (2026-10-23).

On-demand prices from AWS, Google Cloud, Azure, RunPod, Vast.ai, Oracle Cloud, OVHcloud, Scaleway, Verda, Akash and UpCloud, checked every hour. Serving many users needs more KV cache, so more memory: size it in the calculator. Some links are affiliate links: StudioTV may earn a commission, at no extra cost to you.

AWS, Google Cloud and UpCloud prices: Spare Cores (opens in a new tab), CC BY-SA 4.0 (opens in a new tab), shared under the same license.

Recommended settings for MiniMax M3

The repository sets no sampling defaults: these are OpenRouter's defaults for MiniMax M3.

With llama.cpp, pass them yourself: --temp 1 --top-p 0.95

Reasoning. Has a reasoning mode. Reasoning tokens take context and KV cache like any other token: leave room for them.

KV cache: how memory grows with context

Architecture: MoE 128 experts + 1 shared · 60 layers, MiniMax Sparse Attention on 57 (GQA 4 KV × 128, 128-dim index keys) · vision encoder. Its cache stays compact because MLA stores one compressed latent per token and layer instead of full keys and values. Only 23B parameters are active per token, but all 427B must sit in memory: VRAM depends on the total, speed on the active part.

MiniMax M3 VRAM badge

For a model card or a README: the 4-bit size and the smallest GPU it fits on, linked to this page. Data as JSON: /api/models/minimax-m3.json.

[![MiniMax M3 VRAM](https://studiotvai.com/badge/minimax-m3.svg)](https://studiotvai.com/vram-requirements/minimax-m3)

Other models

Same family (large moe): MiniMax M3.1 Flash · MiniMax M2.7 · DeepSeek R1 · Kimi K2.6 · GLM-5.3 · GLM-5.3-Flash · DeepSeek V4 Flash · DeepSeek V4 Pro · DeepSeek V4.1 Flash · MiMo-V2.6 Pro · MiMo-V2.6 Flash · Muse Spark Open · Beam 1.0 501B-A23B · Kimi K3 · Mistral Large 4 · DeepSeek V4 Flash Vision Exp · Qwen3.8 2.4T-A95B · Ornith 1.5 397B · Atria Dawn Preview · Intern S2 397B · DeepSeek V3 · DeepSeek V3.1 · Kimi K2.7 Code · Kimi K2.5 · Kimi K2 · GLM-5.2 · GLM-5.1 · GLM-5 · MiniMax M2.5 · MiniMax M2 · MiMo-V2.5 · Ornith-1.0 397B · Qwen3 235B-A22B · Qwen3.5 397B-A17B · GLM-4.7 · GLM-4.5 · Ling-3.1-flash · Step 5 Preview · Hy3 · Atria-Dawn

Estimates, not guarantees: computed from the official config.json with the same engine as the LLM VRAM Calculator. Real usage depends on your engine version and settings.