RunLocalAI
mistral
675B parameters
Restricted

Mistral Medium 3.5 (675B MoE)

Mistral's April 2026 frontier MoE. 675B total / 41B active. Strong European-multilingual lineage carries through; the new release competes head-to-head with DeepSeek V4-Pro on most benchmarks while landing slightly below it on coding. Mistral Research License — non-commercial use is open; commercial deployment requires a separate Mistral commercial license.

License: Mistral Research License (commercial via separate license)·Released Apr 29, 2026·Context: 262,144 tokens

Overview

Mistral's April 2026 frontier MoE. 675B total / 41B active. Strong European-multilingual lineage carries through; the new release competes head-to-head with DeepSeek V4-Pro on most benchmarks while landing slightly below it on coding. Mistral Research License — non-commercial use is open; commercial deployment requires a separate Mistral commercial license.

Strengths

  • Frontier-class quality with European multilingual depth
  • 41B active params keep tok/s practical on workstation hardware
  • Strong instruction following — Mistral's traditional edge
  • Long context (256K) with verified recall

Weaknesses

  • Research license — commercial use requires Mistral negotiation
  • Behind DeepSeek V4-Pro on coding benchmarks
  • 388 GB at Q4 — workstation cluster required

Quantization variants

Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.

QuantizationFile sizeVRAM required
Q4_K_M388.0 GB448 GB

Get the model

HuggingFace

Original weights

huggingface.co/mistralai/Mistral-Medium-3.5

Source repository — direct quantization required.

Hardware that runs this

Cards with enough VRAM for at least one quantization of Mistral Medium 3.5 (675B MoE).

NVIDIA GB200 NVL72
13824GB · nvidia
Compare alternatives

Models worth comparing

Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.

Step up
More capable — bigger memory footprint
No verdicted models in the next tier up yet.

Frequently asked

What's the minimum VRAM to run Mistral Medium 3.5 (675B MoE)?

448GB of VRAM is enough to run Mistral Medium 3.5 (675B MoE) at the Q4_K_M quantization (file size 388.0 GB). Higher-quality quantizations need more.

Can I use Mistral Medium 3.5 (675B MoE) commercially?

Mistral Medium 3.5 (675B MoE) is released under the Mistral Research License (commercial via separate license), which has restrictions for commercial use. Review the license terms before using it in a product.

What's the context length of Mistral Medium 3.5 (675B MoE)?

Mistral Medium 3.5 (675B MoE) supports a context window of 262,144 tokens (about 262K).

Source: huggingface.co/mistralai/Mistral-Medium-3.5

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.