New B200 spot capacity is live in US East from $1.69 per GPU-hour. See availability →
Deploy a GPU
Mistral AI · Mixture of experts · released December 2025

Mistral Large 3 GPU requirements: 338 GB at 4-bit, 1.35 TB at BF16

Mistral Large 3 needs 1.35 TB of GPU memory for its weights at BF16, 682 GB at FP8 and 338 GB at 4-bit, plus 2.3 GB of KV cache for a 32K context. The cheapest node that runs it here is 8× A40 at 4-bit, for $0.48 an hour on spot.

Architecture read from the model's config.json on Hugging Face, 26 October 2026 · prices from the live catalogue · all 47 GPUs

Cheapest node · 32K context per node-hour, spot
4-bit8× A40 · 340 GB needed $0.48$0.72 on-demand
FP8 (native)8× RTX PRO 6000 Blackwell · 684 GB needed $1.52$3.12 on-demand
BF168× MI300X · 1.35 TB needed $7.92$11.60 on-demand
Deploy 8× A40 All models

Weights plus a 32K 16-bit KV cache, with 10% of the node's memory kept free. Billed per minute.

Parameters
675B41B active per token
Weights, 4-bit
338 GBBF16: 1.35 TB
KV cache, 32K
2.3 GBOne sequence, 16-bit cache
Context window
288KNative
Memory by format

How much VRAM Mistral Large 3 needs, format by format.

Weights are parameters × bytes per parameter, except the native FP8 row, which is the size of the published checkpoint. Each context column adds the KV cache of one sequence of that length.

GPUs by memory
Format Weights + 8Kcontext+ 32Kcontext+ 128Kcontext+ 288Kcontext Cheapest node, 32K
BF16Published weights, fine-tuning 1.35 TB 1.35 TB1.35 TB1.36 TB1.37 TB 8× MI300X · $7.92/h
FP8 (native)Hopper, Ada, Blackwell, MI300 682 GB 682 GB684 GB691 GB702 GB 8× RTX PRO 6000 Blackwell · $1.52/h
4-bitAWQ, GPTQ, GGUF Q4 338 GB 338 GB340 GB347 GB358 GB 8× A40 · $0.48/h
Arithmetic on published facts, not a benchmark.

The serving engine adds its own buffers, and an 8-bit KV cache halves the context columns. We keep 10% of a node's memory free when we say a model fits. Speed is a separate question: it follows memory bandwidth and, for a mixture of experts, the active parameters.

Where it runs

GPUs that run Mistral Large 3, cheapest first.

For each format, the smallest node of each GPU model that holds the weights and a 32K context, sorted by spot price. One row per GPU model; larger nodes of the same GPU work too.

All prices
Format GPU GPU memory Spot On-demand 730 h spot
4-bit 8× A40Ampere · 340 GB needed 384 GB $0.48 $0.72 $350.40 Deploy
4-bit 8× L40Ada Lovelace · 340 GB needed 384 GB $0.72 $1.52 $525.60 Deploy
4-bit 8× A100 PCIe 80GBAmpere · 340 GB needed 640 GB $0.72 $1.52 $525.60 Deploy
4-bit 4× RTX PRO 6000 BlackwellBlackwell · 340 GB needed 384 GB $0.76 $1.56 $554.80 Deploy
FP8 (native) 8× RTX PRO 6000 BlackwellBlackwell · 684 GB needed 768 GB $1.52 $3.12 $1,109.60 Deploy
FP8 (native) 4× MI300XCDNA 3 · 684 GB needed 768 GB $3.96 $5.80 $2,890.80 Deploy
FP8 (native) 4× MI325XCDNA 3 · 684 GB needed 1024 GB $4.36 $6.36 $3,182.80 Deploy
FP8 (native) 8× H200 NVLHopper · 684 GB needed 1128 GB $5.52 $7.92 $4,029.60 Deploy
BF16 8× MI300XCDNA 3 · 1.35 TB needed 1536 GB $7.92 $11.60 $5,781.60 Deploy
BF16 8× MI325XCDNA 3 · 1.35 TB needed 2048 GB $8.72 $12.72 $6,365.60 Deploy

Spot capacity can be reclaimed with a 2-minute notice; for a live endpoint, use on-demand. How spot works · serving a language model.

Context and the KV cache

What a long context costs in memory.

Multi-head latent attention: one compressed 576-value vector per token and layer, 61 layers.

Context KV cache, 1 sequence 8 sequences
8K tokens 0.58 GB 4.61 GB
32K tokens 2.3 GB 18.4 GB
128K tokens 9.21 GB 73.7 GB
288K tokens 20.7 GB 166 GB

Serving several users

KV cache

Every concurrent request keeps its own cache, so memory for context grows with users × context length. A server batching 8 requests of 32K needs eight times the single-sequence figure — 18.4 GB here.

  • Recommended node8× A40
  • Weights at 4-bit338 GB
  • Context that fits, 1 sequence113K

Paged attention (vLLM, SGLang) allocates the cache in blocks as requests grow, so a node holds more short requests than the worst case suggests.

Model facts

Mistral Large 3, as published.

Developer
Mistral AIMistral flagship mixture of experts
Parameters
675B total, 41B active per token
Architecture
Mixture of experts, 61 layersMulti-head latent attention: one compressed 576-value vector per token and layer, 61 layers
Context window
294,912 tokensAs set in the published configuration
Published format
FP8, 682 GB checkpoint
Licence
Apache 2.0
Released
2 December 2025
Source
mistralai/Mistral-Large-3-675B-Instruct-2512config.json read 26 October 2026

Serve it with vLLM

template: vLLM

On 8× A40, from the instance terminal:

terminal
$ vllm serve mistralai/Mistral-Large-3-675B-Instruct-2512 \
    --tensor-parallel-size 8 \
    --max-model-len 32768

Very recent architectures need a recent vLLM; the model card gives the minimum version and any extra flags.

FAQ

Mistral Large 3 on a GPU — the questions that come up

How much VRAM does Mistral Large 3 need?

For the weights alone: 1.35 TB at BF16, 682 GB at FP8 and 338 GB at 4-bit. Add 2.3 GB of KV cache for one 32K-token context, or 20.7 GB at the full 288K window, and keep about 10% of the memory free for activations.

Can Mistral Large 3 run on a single GPU?

One RTX 4090 (24 GB): no. One RTX 5090 (32 GB): no. One H100 SXM (80 GB): no. One H200 SXM (141 GB): no. One B200 SXM (180 GB): no. Those answers assume a 32K context and the widest format that still fits.

What is the cheapest way to run Mistral Large 3?

Here, 8× A40 at 4-bit: $0.48 per hour on spot and $0.72 on-demand, billed per minute — $350.40 for a 730-hour month on spot. That node has 384 GB of GPU memory for the 340 GB this setup needs.

Does Mistral Large 3 need less memory because only 41B parameters are active?

No. A mixture-of-experts model routes each token through a few experts, but every expert has to be resident in GPU memory, so the memory bill follows the 675B total. What the 41B active parameters change is the work per token: generation reads far less of the weights each step than a dense model of the same size would, which is why it is fast for its size.

Which format should I run it in?

Mistral Large 3 is published in FP8, a 682 GB checkpoint — the format its developers serve it in, so it is the safe default. BF16 keeps the published weights untouched and is the format for fine-tuning. FP8 halves the memory and runs in hardware on Hopper, Ada, Blackwell and MI300-class GPUs. 4-bit quantisation (AWQ, GPTQ, GGUF) quarters it, at a quality cost that is larger on small models than on big ones; check it against your own evaluation before relying on it.

How do I deploy it?

Pick the node in the table above and launch it with the vLLM template; the model downloads from Hugging Face (mistralai/Mistral-Large-3-675B-Instruct-2512) on first start. Spread the weights over the node with --tensor-parallel-size equal to the GPU count, and set --max-model-len to the context you need. New architectures need a recent vLLM: check the model card for the minimum version.

Sizing another model? GPU requirements for 64 open models · serving a language model · fine-tuning.

Get started

Mistral Large 3 on 8× A40 for $0.48 an hour.

Launch the node with the vLLM template, point it at mistralai/Mistral-Large-3-675B-Instruct-2512, stop it when you are done. Prepaid, billed by the minute.

Deploy 8× A40 All open models

Billed per minute from the moment the instance is reachable. Minimum credit $30, no subscription.