llmrun
Mistral AIMistralMistralForCausalLM

Magistral Small 2506 — Hardware Requirements & GPU Compatibility

Chat

Magistral Small 2506 is Mistral AI's small, efficient reasoning model, built on the 24-billion-parameter Mistral Small 3.1 and further trained with supervised fine-tuning on Magistral Medium's reasoning traces plus reinforcement learning. It produces long chains of reasoning before answering and supports dozens of languages, making it suited to multi-step problem solving rather than simple chat. The architecture supports up to 128K tokens, but Mistral recommends keeping usage around 40K since output quality can degrade past that point. Released under the Apache 2.0 license, it is compact enough to fit on a single RTX 4090 or a 32 GB RAM Mac once quantized to 4-bit, in line with what a 24-billion-parameter model needs at that precision.

23.6B params · dense41K contextApache 2.070.3K downloads617 likes13.7K quant downloads
Based on Mistral Small 3.1 24B Instruct 2503
A

Runs well on a 24 GB GPU

Runs from12 GB IQ2_M · 16KBest on24 GB Q4_K_M · ~44 tok/s · 40KMacs from16 GB IQ2_M · ~16 tok/s

Score 88.7#70 of 100 open#19 of 35 that fit 24 GBApache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

llmrun Composite Score →

88.7

Benchmarks · tier chart ↓
Among 100 open modelsHow the Composite Score is built

Run Magistral Small 2506

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Recommended

Ollama
one command · needs Ollama installed
$ollama run hf.co/bartowski/mistralai_Magistral-Small-2506-GGUF:Q4_K_M

Q4_K_M fits fully on the RTX 4090 · 6.2 GB free · 16K context

Download: mistralai_Magistral-Small-2506-Q4_K_M.gguf14.33 GBbartowski

install, OS, other quants

Other ways

each opens step-by-step

Run in cloud

Fits on RTX 3090 (24 GB) (6 GB headroom) · Q4_K_M

Generation speed
~41 tok/s
generation speed
Cost per 1M output tokens
$1.37
per 1M output tokens
Deployon Vast.aiBest optionfrom $0.20/hr (opens in new tab)
Compare GPUs →
or

How Much VRAM Does Magistral Small 2506 Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KOn RTX 4090 (reference)
10.7 GB17 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsFits: 40K context · ~61 tok/s
11.0 GB17.3 GB3-bit small — 3-bit small quantizationFits: 40K context · ~59 tok/s
12.2 GB18.5 GB3-bit medium — 3-bit medium quantizationFits: 40K context · ~54 tok/s
12.5 GB18.8 GB4-bit · legacy — 4-bit legacy quantizationFits: 40K context · ~52 tok/s
14.9 GB21.2 GB4-bit medium — 4-bit medium quantization — most popular sweet spotFits: 40K context · ~44 tok/s
17.5 GB23.8 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffFits: 32K context · ~37 tok/s
20.2 GB26.5 GB6-bit, very good — 6-bit quantization, very good qualityFits: 16K context · ~33 tok/s
24.3 GB30.6 GB8-bit, near-lossless — 8-bit quantization, near-losslessDoes not fit: needs 27.2 GB

Selected Q4_K_M

4-bit medium quantization

17.8GBat 16K context

Weights
14.1 GB
KV cache
3.4 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 6.2 GB free · ~44 tok/s

≥50 tok/s needs ≥1.14 TB/s · RTX 5090 ~78

Which Hardware Can Run Magistral Small 2506?

Q4_K_M · 14.9 GB

Magistral Small 2506 (Q4_K_M) requires 14.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 20+ GB is recommended. Using the full 41K context window can add up to 8.0 GB, bringing total usage to 22.8 GB. 26 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080. 122 devices with unified memory can run Magistral Small 2506, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M5 Pro (24 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M
8 GB · RTX 4060Doesn't fit — smallest quant needs 10.1 GB8 GB tier →
12 GB · RTX 3060 12GBIQ2_M · 11.6 GB · ~27 tok/s · 16K12 GB tier →
16 GB · RTX 5060 Ti 16GBQ4_K_M · 17.8 GB · ~20 tok/s · 4Ktight16 GB tier →
24 GB · RTX 4090referenceQ4_K_M · 17.8 GB · ~44 tok/s · 40K24 GB tier →
Mac 32 GB · Mac Mini M4Q4_K_M · 17.8 GB · ~6 tok/s · 40KMacs →
Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 17.8 GB · ~26 tok/s · 40KMacs →

How good is Magistral Small 2506?

The llmrun Score merges 4 public benchmark boards into a single figure, where 100 represents the average for open models.

Score vs VRAM in the 24 GB tier

  • Magistral Small 2506
  • Other models
  • Best Score for its VRAM

#19 of 35 that fit 24 GB

Each dot is a model at the best quant that fits a 24 GB GPU. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full 24 GB ranking →

Benchmarks

Here are all 4 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.

  • GPQA Diamond

    Epoch AI · Jun 2025Independent

    56.1%

    #47/98 open

    Independent
  • AIME 2024/2025

    Epoch AI · Jun 2025Independent

    30.0%

    #50/86 open

    Independent
  • ARC-AGI-2

    Epoch AI · Jun 2025Independent

    0.0%

    #19/19 open

    Independent
  • ARC-AGI

    Epoch AI · Jun 2025Independent

    5.0%

    #17/19 open

    Independent
All 4 Magistral Small 2506 benchmarks

In the 24 GB class

35 models fit 24 GB

Same Score, only models that fit 24 GB at their best quant: the ranking a 24 GB owner actually chooses from.

  1. #1

    Qwen3.8 27BAlibaba

    122.1

    ~38 tok/s · 17.9 GB

  2. #2

    Qwen3.6 27BAlibaba

    111.2

    ~38 tok/s · 17.9 GB

  3. #3

    Gemma 4 31B ITGoogle

    106.8

    ~32 tok/s · 21.7 GB

  4. #4

    Qwen3.6 35B A3BAlibaba

    106.8

    ~151 tok/s · 22.0 GB

  5. #5

    Gemma 4 26B A4B ITGoogle

    105.8

    ~171 tok/s · 16.5 GB

  6. #6

    Qwen3.5 35B A3BAlibaba

    105.8

    ~151 tok/s · 22.0 GB

Full 24 GB ranking →RTX 4090 page →

Alternatives to Magistral Small 2506

Mistral family · 34 models →

Official models with an llmrun Score, each answering one question; numbers are at the best quant that fits a 24 GB GPU.

Similar models

10
ModelScoreVRAMtok/sWhy it is similar
Gemma 4 31B ITGoogle106.821.7 GB~32similar score
Gemma 4 26B A4B ITGoogle105.816.5 GB~171similar score
Qwen3.5 35B A3BAlibaba105.822 GB~151similar score
Qwen3.5 27BAlibaba103.817.9 GB~38similar score
Seed OSS 36B InstructByteDance-Seed102.522.8 GB~31similar score
Qwen3 30B A3B Thinking 2507Alibaba100.719.4 GB~173similar score
Qwen3.5 9BAlibaba99.46.6 GB~103similar score
Qwen3 32BAlibaba98.622.6 GB~32similar score
Qwen3 14BAlibaba98.011.9 GB~69similar score
DeepSeek R1 Distill Qwen 32BDeepSeek97.223 GB~34similar score

Specifications

Publisher
Mistral AI
Family
Mistral
Parameters
23.6B
Architecture
MistralForCausalLM
Context Length
40,960 tokens
Vocabulary Size
131,072
Release Date
2025-06-04
License
Apache 2.0

Resources

Frequently Asked Questions

How much VRAM does Magistral Small 2506 need?

Magistral Small 2506 requires 14.9 GB of VRAM at Q4_K_M, or 47.9 GB at BF16. Full 41K context adds up to 8.0 GB (22.8 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 23.6B × 4.8 bits ÷ 8 = 14.1 GB

KV Cache + Overhead ≈ 0.8 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 8.7 GB (at full 41K context)

VRAM usage by quantization

14.9 GB
22.8 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 4090 run Magistral Small 2506?

Yes, at Q6_K (20.2 GB) or lower. Higher quantizations like Q8_0 (24.3 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.

What's the best quantization for Magistral Small 2506?

For Magistral Small 2506, Q4_K_M (14.9 GB) offers the best balance of quality and VRAM usage. Q4_K_L (15.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 7.2 GB.

VRAM requirement by quantization

IQ2_XXS
7.2 GB
IQ3_XS
10.4 GB
Q3_K_L
12.8 GB
Q4_K_M ★
14.9 GB
Q4_K_L
15.2 GB
BF16
47.9 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Magistral Small 2506 on a Mac?

Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Magistral Small 2506, the smallest at IQ2_M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 14.9 GB of usable unified memory (RAM minus macOS overhead), from Mac Mini M5 Pro (24 GB).

Can I run Magistral Small 2506 locally?

Yes — Magistral Small 2506 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Magistral Small 2506?

At Q4_K_M, Magistral Small 2506 can reach ~323 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~44 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 14.86 × 0.65 = ~350 tok/s

Estimated speed at Q4_K_M (14.9 GB)

~350 tok/s
~44 tok/s
~350 tok/s
~323 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Magistral Small 2506?

At Q4_K_M, the download is about 14.14 GB. The full-precision BF16 version is 47.14 GB. The smallest option (IQ2_XXS) is 6.48 GB.

Which GPUs can run Magistral Small 2506?

26 consumer GPUs can run Magistral Small 2506 at Q4_K_M. Top options include AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, NVIDIA GeForce RTX 3090 Ti, AMD Radeon RX 6800. 7 GPUs have plenty of headroom for comfortable inference.

Which devices can run Magistral Small 2506?

122 devices with unified memory can run Magistral Small 2506 at Q4_K_M (14.9 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.