How much VRAM does Mistral Large 3 (MoE) need?

Mistral Large 3 (MoE) requires approximately 0.0 GB VRAM at Q4_K_M quantization with 1K context and 10% overhead. At Q8_0 it needs ~0.0 GB, and at FP16 ~0.0 GB. Use the PCPARTGUIDE VRAM Calculator for exact requirements at any context length and overhead setting.

Can I run Mistral Large 3 (MoE) on a consumer GPU?

Mistral Large 3 (MoE) is a 675B parameter model and requires datacenter or multi-GPU setups. At Q4_K_M, it needs approximately 0.0 GB VRAM, which exceeds consumer GPU limits. It can run on H100 (80 GB) or A100 (80 GB) datacenter GPUs, or across multiple consumer GPUs with tensor parallelism.

What is Mistral Large 3 (MoE)'s context window?

Mistral Large 3 (MoE) supports a maximum context window of 262,144 tokens. The default context is 4,096 tokens. Larger contexts require proportionally more VRAM for the KV cache — use the VRAM Calculator to estimate exact requirements at your target context length.

What quantization level should I use for Mistral Large 3 (MoE)?

For Mistral Large 3 (MoE), start with Q4_K_M (~0.0 GB) if VRAM-limited. Q8_0 offers better quality at ~2× the VRAM. If you have enough VRAM, Q6_K (~1.5× Q4 VRAM) provides a good balance. For models this large, Q3_K_M may be necessary to fit on consumer hardware.

How does Mistral Large 3 (MoE) compare to similar models?

Mistral Large 3 (MoE) is a 675.0B MoE model from the Mistral family. Browse other Mistral models on the PCPARTGUIDE directory for comparable alternatives with different parameter counts and VRAM requirements. Use the Model Transparency scores to assess documentation quality and reproducibility.

MistralMoEApache 2.0

Mistral Large 3 (MoE)

675.0B

Parameters

41.0B

Active

256K

Max Context

MoE

Architecture

—

Released

Text

Modality

About Mistral Large 3 (MoE)

Mistral Large 3 (MoE) is a mixture-of-experts (MoE) transformer language model from the Mistral family, containing 675B parameters across 88 layers. It has 675B total parameters loaded into VRAM with 41B active per token. It supports up to 262K tokens of context with a hidden dimension of 12288 and 8 KV heads for efficient grouped-query attention (GQA). Apache 2.0. MoE: 128 experts, top-4 routing. Server class.

ResearchEnterprise

Technical Specifications

Total Parameters675.0B

Active Parameters41.0B per token

ArchitectureMixture of Experts

Total Experts41

Attention TypeGQA (MoE)

Hidden Dimensiond = 12,288

Transformer Layers88

Attention Heads96

KV Headsn_kv = 8

Head Dimensiond_head = 128

Activation FunctionSwiGLU

NormalizationRMSNorm

Position EmbeddingRoPE

System Requirements

Estimated VRAM at 10% overhead for different quantization methods and context sizes.

Quantization	1K ctx	195K ctx	256K ctx
Q4_K_M0.50 B/W ~97% of FP16	349.2Cluster / Multi-GPU	416.0Cluster / Multi-GPU	436.9Cluster / Multi-GPU
Q8_01.00 B/W ~100% of FP16	698.1Cluster / Multi-GPU	764.9Cluster / Multi-GPU	785.8Cluster / Multi-GPU
F162.00 B/W Reference	1395.9Cluster / Multi-GPU	1462.7Cluster / Multi-GPU	1483.6Cluster / Multi-GPU

Use the interactive VRAM Calculator to see exactly how much memory you need at any quantization level, context length, and overhead setting.

Open VRAM Calculator Best GPU Guide VRAM Guide