LLMBOTTLENECK.COM
Mistral AI / architecture

Mistral-Large-3-675B-Instruct-2512

675B total · 673B language model · 41B active, routing 4 of 128 experts per token.

The official Mistral card declares 675B total and 41B active for the whole model, and separates it into "A Granular MoE Language Model with 673B params and 39B active" and "A 2.5B Vision Encoder". The 673B language model is what this engine sizes; the vision encoder is a separate tower it does not model. Hugging Face publishes no safetensors element total for this repository — it ships Mistral's reference format, with consolidated-*.safetensors and no Transformers index — so the card's own figure is what the pinned architecture is reconciled against. Open the official parameter note ↗

The official Mistral card documents a 256k context window; params.json states max_position_embeddings 294912 and one shared expert. Open the official context documentation ↗

Open in the calculator →

Architecturepublished data
Quick answer
Mistral-Large-3-675B-Instruct-2512 needs about 404.1 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. None of the 18 common devices listed below holds it entirely at this setting. No hardware for it? Rent a GPU that runs it, priced live.

Sized at 8,192 tokens of context with the whole model in device memory, across 18 common devices. Speeds are estimates, not benchmarks.

Published in Mistral's reference format: the geometry is in params.json, and this repository ships no Transformers config.json. Sized from the text decoder; the params.json vision_encoder block is a separate image tower this engine does not size. Licence declared on the repository: apache-2.0.

Weight-file size by format

FormatSizeRangeBasis
FP161336.7 GB1320.7 GB – 1352.7 GBreconstructed size
Q8_0710.2 GB701.7 GB – 718.7 GBreconstructed size
Q6_K548.4 GB541.8 GB – 555.0 GBreconstructed size
Q5_K_M473.4 GB467.7 GB – 479.0 GBreconstructed size
Q5_0459.8 GB454.3 GB – 465.3 GBreconstructed size
Q4_K_M402.7 GB397.9 GB – 407.6 GBreconstructed size
Q4_0376.4 GB371.9 GB – 380.9 GBreconstructed size
Q3_K_M329.7 GB304.3 GB – 355.0 GBreconstructed size
Q2_K242.2 GB230.9 GB – 253.6 GBreconstructed size

A file size is not the memory a run needs: the KV cache and the runtime reserve come on top, and the calculator adds both.

What the labels mean
published data
Read from a published source — a file's byte count, a model's configuration or a manufacturer's specification — or exact arithmetic on such values. Not a measurement on a machine.
reconstructed size
Weight size reconstructed from the pinned architecture, because no published file exists.
size range
Only a lower and an upper bound are claimed for this weight size.

Hardware ladder

Q4_K_M at 8,192 tokens · 0 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 5070404.1 GB of 12 GB60 layers on system RAM~3.0 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT404.1 GB of 16 GB59 layers on system RAM~3.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 4080404.1 GB of 16 GB59 layers on system RAM~3.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5080404.1 GB of 16 GB59 layers on system RAM~3.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5070 Ti404.1 GB of 16 GB59 layers on system RAM~3.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5060 Ti 16GB404.1 GB of 16 GB59 layers on system RAM~3.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 4090404.1 GB of 24 GB58 layers on system RAM~3.1 tok/s with offloadOpen →
AMD Radeon™ RX 7900 XTX404.1 GB of 24 GB58 layers on system RAM~3.1 tok/s with offloadOpen →
NVIDIA GeForce RTX 3090404.1 GB of 24 GB58 layers on system RAM~3.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5090404.1 GB of 32 GB57 layers on system RAM~3.1 tok/s with offloadOpen →
NVIDIA RTX 6000 Ada Generation404.1 GB of 48 GB54 layers on system RAM~3.3 tok/s with offloadOpen →
Apple M4 Pro404.1 GB of 64 GBdoes not fitnot calibratedOpen →
Apple M5 Pro404.1 GB of 64 GBdoes not fitnot calibratedOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S404.1 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M3 Max404.1 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M4 Max404.1 GB of 128 GBdoes not fitnot calibratedOpen →
NVIDIA DGX Spark404.1 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M2 Ultra404.1 GB of 192 GBdoes not fitnot calibratedOpen →

Decode is a calibrated estimate from the published calibration; capacity is the manufacturer's published ceiling, not guaranteed free memory. Fitting in memory is not the same as loading: whether the runtime and version you have supports this architecture and format on that machine has not been tested here. Offloaded rows assume there is enough system RAM for the overflow — the calculator checks that against the RAM you declare. Each “Open” link keeps this model, format and context.

Run it in the cloud

Rent a GPU that runs Mistral-Large-3-675B-Instruct-2512

Every rentable machine that holds the whole model, from Vast.ai and RunPod, with this site's speed estimate for each and the provider's own price, read live. Cheapest first; sort by cost per token or speed instead, or switch the format.

Prices · 18:30 UTC, 22 Sept
6 configurations in stock · 5 out of stock
MachineSpeedPer hourPer M tokensWhereRent
2× B300540 GBcheapest~482 tok/sFaster than you read$15.78$9.09RunPodSecure CloudRent on RunPod
4× H200 NVL564 GB~571 tok/sFaster than you read$16.00$7.78Vast.aiMarketplace · 99.1% reliableRent on Vast.ai
8× RTX PRO 6000768 GB~387 tok/sFaster than you read$16.54$11.86Vast.aiMarketplace · 98.4% reliableRent on Vast.ai
4× H200564 GB~571 tok/sFaster than you read$18.36$8.93RunPodSecure CloudVast.ai $18.42/h Rent on RunPod
8× H100 PCIe640 GB~432 tok/sFaster than you read$19.16$12.31Vast.aiMarketplace · 98.3% reliableRent on Vast.ai
4× B200720 GBrecommendedbest valuefastest~916 tok/sFaster than you read$25.00$7.58Vast.aiMarketplace · 99.6% reliableRent on Vast.ai
First time renting a GPU? How it works, in four steps
  1. Create an account and add credit. Both providers are prepaid: $10 is enough to try any card on this page for hours. A new RunPod account that arrives through a referral link gets a one-time $5 credit when it first adds $10.
  2. Pick the card and the count shown here, and a template: llama.cpp or vLLM for an OpenAI-compatible API, or one with a web chat if you only want to talk to the model. “How to launch it” below gives the exact image and command.
  3. Wait for the download. The weights are fetched on the machine; tens of gigabytes take a few minutes on a data-centre connection. You pay by the second from the moment it starts.
  4. Stop it when you are done. A running machine bills even when idle. A stopped one costs nothing per hour, though its disk is usually still billed until you delete it.
How to launch it

No Q4_K_M file of Mistral-Large-3-675B-Instruct-2512 is catalogued here, so there is no exact command to vouch for. Start the machine from the llama.cpp server image ghcr.io/ggml-org/llama.cpp:server-cuda and point -hf at a Q4_K_M build from the publisher or a quantizer you trust — search Hugging Face for one. Check its file size against the memory figure on this page before you rent.

Every machine holds the whole model at Q4_K_M and 8,192 tokens of context, no offload. Speeds are this site’s single-stream decode estimates; prices are what each provider’s own API quoted, on-demand, for the whole machine. Vast hosts below 98% measured reliability are left out.

Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank

Buying a card instead, or paying by the token? Compare a year of Mistral-Large-3-675B-Instruct-2512 three ways →

Under the hood

Architecture, read from the publisher’s file

The numbers every figure above is computed from, with the file they came from.

Architecture

✓ Architecture read from the published config.json

Retrieved 2026-09-20 at pinned commit 383ffea2c7d6.

Layers
61
Hidden size
7,168
Attention heads
128
KV heads
128
Head dimension
192
Feed-forward width
16,384
Vocabulary
131,072
Context ceiling
262,144
Config max_position_embeddings
294,912
RoPE theta
10,000
Experts
128
Experts / token
4
Expert width
4,096
Shared experts
1
Latent KV rank
512

Where the memory goes

tokenembedding× 61 decoder blocksLatent attentioncompressed KV rankfull contextRouted experts4 of 128 per tokenall residentoutputprojectiongrows with contextcapacity ≠ traffic

Latent attention projects keys and values into a compressed rank before caching them, which is why its KV cache is a fraction of a comparable dense model.

Open the exact official config.json ↗

More from Mistral AI

Devstral-2-123B-Instruct-2512Devstral-Small-2507Magistral-Small-2509Ministral-3-14B-Instruct-2512Ministral-3-3B-Instruct-2512Ministral-3-8B-Instruct-2512Mistral-7B-Instruct-v0.2Mistral-7B-Instruct-v0.3

Use this answer in your own product

As JSON, from the API
curl -s https://llmbottleneck.com/v1/analyze \
  -H "Authorization: Bearer $LLMB_KEY" -H "content-type: application/json" \
  -d '{"model":"mistralai-mistral-large-3-675b-instruct-2512","quantization":"Q4_K_M","context":8192,"hardware":"nvidia-geforce-rtx-4090"}'

Same engine, same evidence, every field sourced. Free key in one step →

On your page, as a widget
<script src="https://llmbottleneck.com/widget.js"
  data-model="mistralai-mistral-large-3-675b-instruct-2512" data-quantization="Q4_K_M"
  data-hardware="nvidia-geforce-rtx-4090" data-context="8192"></script>

No key needed. Unbranded, with your own buy button, on Pro and Business →

Citing this page

LLM Bottleneck. “Mistral-Large-3-675B-Instruct-2512 VRAM and hardware requirements.” Architecture from mistralai/Mistral-Large-3-675B-Instruct-2512 at revision 383ffea2c7d6, retrieved 2026-09-20. https://llmbottleneck.com/models/mistralai-mistral-large-3-675b-instruct-2512

Every figure above is either the published value or a reconstruction whose measured error is on the accuracy page.

llmbottleneck
catalogue 2026-10-03models 327devices 135