LLMBOTTLENECK.COM
Mistral AI / mistral

Devstral-Small-2507

23.6 billion parameters.

Open in the calculator →

Architecturepublished data
Quick answer
Devstral-Small-2507 needs about 16.5 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. It fits entirely on 12 of 18 common devices, starting with the NVIDIA GeForce RTX 4090 (24 GB). What else fits in 24 GB. 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.

Find the NVIDIA GeForce RTX 4090, the smallest card listed here that holds it: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

Weight-file size by format

FormatSizeRangeBasis
FP1647.2 GBexactpublished data
Q8_025.1 GBexactpublished data
Q6_K19.3 GB18.8 GB – 19.4 GBreconstructed size
Q5_K_M16.8 GBexactpublished data
Q5_016.3 GB15.7 GB – 16.4 GBreconstructed size
Q4_K_M14.3 GBexactpublished data
Q4_013.4 GB12.9 GB – 13.6 GBreconstructed size
Q3_K_M11.7 GB10.9 GB – 12.6 GBreconstructed size
Q2_K8.77 GB8.22 GB – 9.13 GBreconstructed size

4 of these are published files. Open the pinned GGUF repository ↗

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 · 12 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 507016.5 GB of 12 GB13 layers on system RAM~10 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT16.5 GB of 16 GB2 layers on system RAM~24 tok/s with offloadOpen →
NVIDIA GeForce RTX 408016.5 GB of 16 GB2 layers on system RAM~24 tok/s with offloadOpen →
NVIDIA GeForce RTX 508016.5 GB of 16 GB2 layers on system RAM~29 tok/s with offloadOpen →
NVIDIA GeForce RTX 5070 Ti16.5 GB of 16 GB2 layers on system RAM~28 tok/s with offloadOpen →
NVIDIA GeForce RTX 5060 Ti 16GB16.5 GB of 16 GB2 layers on system RAM~17 tok/s with offloadOpen →
NVIDIA GeForce RTX 409016.5 GB of 24 GBfits~45 tok/sOpen →
AMD Radeon™ RX 7900 XTX16.5 GB of 24 GBfits~49 tok/sOpen →
NVIDIA GeForce RTX 309016.5 GB of 24 GBfits~42 tok/sOpen →
NVIDIA GeForce RTX 509016.5 GB of 32 GBfits~80 tok/sOpen →
NVIDIA RTX 6000 Ada Generation16.5 GB of 48 GBfits~43 tok/sOpen →
Apple M4 Pro16.5 GB of 64 GBfits~15 tok/sAbout reading paceOpen →
Apple M5 Pro16.5 GB of 64 GBfits~17 tok/sAbout reading paceOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S16.5 GB of 128 GBfits~13 tok/sAbout reading paceOpen →
Apple M3 Max16.5 GB of 128 GBfits~21 tok/sAbout reading paceOpen →
Apple M4 Max16.5 GB of 128 GBfits~28 tok/sAbout reading paceOpen →
NVIDIA DGX Spark16.5 GB of 128 GBfits~12 tok/sAbout reading paceOpen →
Apple M2 Ultra16.5 GB of 192 GBfits~38 tok/sOpen →

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 Devstral-Small-2507

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
27 configurations in stock · 6 out of stock
MachineSpeedPer hourPer M tokensWhereRent
2× RTX 306024 GBcheapest~32 tok/sFaster than you read$0.11$0.95Vast.aiMarketplace · 98.0% reliableRent on Vast.ai
1× RTX 309024 GBrecommendedbest value~41 tok/sFaster than you read$0.12$0.82Vast.aiMarketplace · 99.5% reliableRunPod $0.50/h Rent on Vast.ai
1× RTX 3090 Ti24 GB~44 tok/sFaster than you read$0.19$1.16Vast.aiMarketplace · 99.6% reliableRunPod $0.27/h Rent on Vast.ai
2× RTX 407024 GB~44 tok/sFaster than you read$0.19$1.17Vast.aiMarketplace · 99.8% reliableRent on Vast.ai
2× RTX 4060 Ti 16GB32 GB~25 tok/sAbout reading pace$0.22$2.43Vast.aiMarketplace · 99.7% reliableRent on Vast.ai
1× L424 GB~13 tok/sAbout reading pace$0.26$5.34Vast.aiMarketplace · 98.9% reliableRunPod $0.49/h Rent on Vast.ai
2× RTX 5060 Ti 16GB32 GB~39 tok/sFaster than you read$0.26$1.78Vast.aiMarketplace · 99.5% reliableRent on Vast.ai
1× RTX 6000 Ada48 GB~42 tok/sFaster than you read$0.26$1.70Vast.aiMarketplace · 99.9% reliableRunPod $0.84/h Rent 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 Q4_K_M on it (llama.cpp)
Container image
ghcr.io/ggml-org/llama.cpp:server-cuda
Start command / arguments
-hf mistralai/Devstral-Small-2507_gguf:Q4_K_M -c 8192 -ngl 999 --host 0.0.0.0 --port 8080
Expose port
8080 — an OpenAI-compatible API at /v1

Starting from a saved llama.cpp template instead? Leave the arguments empty and set these variables — only the first two change between models:

LLAMA_ARG_HF_REPO=mistralai/Devstral-Small-2507_gguf:Q4_K_M
LLAMA_ARG_CTX_SIZE=8192
LLAMA_ARG_N_GPU_LAYERS=999
LLAMA_ARG_HOST=0.0.0.0
LLAMA_ARG_PORT=8080

Already have the runtime? llama-server -hf mistralai/Devstral-Small-2507_gguf:Q4_K_M -c 8192 -ngl 999 --host 0.0.0.0 --port 8080. Weights: mistralai/Devstral-Small-2507_gguf. The CUDA image is for NVIDIA cards; an AMD machine needs llama.cpp's ROCm build.

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 Devstral-Small-2507 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 2026-09-01 at pinned commit bd165ab26ceb; parameter count from the safetensors index at the same revision.

Architecture
mistral
Layers
40
Hidden size
5,120
Attention heads
32
KV heads
8
Head dimension
128
Feed-forward width
32,768
Vocabulary
131,072
Context ceiling
131,072
RoPE theta
1,000,000,000

Where the memory goes

tokenembedding× 40 decoder blocksGrouped-query attention32 query · 8 KV headsfull contextFeed-forwardone networkall activeoutputprojectiongrows with contextfixed per token

Grouped-query attention shares each key/value head across 4 query heads, so the KV cache is 25% of what multi-head attention would need at the same context.

Open the exact official config.json ↗

More from Mistral AI

Devstral-2-123B-Instruct-2512Magistral-Small-2509Ministral-3-14B-Instruct-2512Ministral-3-3B-Instruct-2512Ministral-3-8B-Instruct-2512Mistral-7B-Instruct-v0.2Mistral-7B-Instruct-v0.3Mistral-7B-v0.1

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-devstral-small-2507","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-devstral-small-2507" 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. “Devstral-Small-2507 VRAM and hardware requirements.” Architecture from mistralai/Devstral-Small-2507 at revision bd165ab26ceb, retrieved 2026-09-01. https://llmbottleneck.com/models/mistralai-devstral-small-2507

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