Magistral Small 2506 vs Qwen3.8 27B

Qwen3.8 27B leads llmrun Score 121.5 to 88.7; both fit 24 GB at Q4_K_M.

Model A · Mistral AI

Magistral Small 2506

88.7llmrun Score

23.6B · 40K context · Apache 2.0 · Jun 2025

Chat

Model B · Alibaba

Qwen3.8 27B

121.5llmrun Score

27.8B · 256K context · Apache 2.0 · Aug 2026

VisionFunctions

Which should you choose?

Choose Magistral Small 2506 if you want…

Also does chat
Qwen3.8 27B doesn't list it

Choose Qwen3.8 27B if you want…

Higher llmrun Score
121.5 vs 88.7
Stronger at reasoning
124.1 vs 88.2
Fits a 12 GB GPU
IQ3_XXS · 12 GB; Magistral Small 2506 needs 10.1 GB at its smallest quant
Longer context window
256K vs 40K tokens
Also does vision, function-calling
Magistral Small 2506 doesn't list it
Newer release
Aug 2026 vs Jun 2025

On your hardware

Each model at its default quant (Q4_K_M), a smaller one when it must. Estimates, ±20%.

MemoryMagistral Small 2506Qwen3.8 27B

8 GB

RTX 4060 · 8 GB tier

Magistral Small 2506

Doesn't fit

smallest quant needs 10.1 GB

Qwen3.8 27B

Doesn't fit

smallest quant needs 8.8 GB

12 GB

RTX 3060 12GB · 12 GB tier

Magistral Small 2506

IQ2_M · ~27 tok/s

11.6 GB · up to 16K

Qwen3.8 27B

IQ3_XXS · ~20 tok/s

12 GB · up to 16Kbetter quant

16 GB

RTX 5060 Ti 16GB · 16 GB tier

Magistral Small 2506

Q4_K_M · ~20 tok/stight

17.8 GB · up to 4K

Qwen3.8 27B

Q3_K_L · ~19 tok/s

15.4 GB · up to 16Kmore headroom

24 GBreference

RTX 4090 · 24 GB tier

Magistral Small 2506

Q4_K_M · ~44 tok/s

17.8 GB · up to 40Kfaster

Qwen3.8 27B

Q4_K_M · ~38 tok/s

17.9 GB · up to 64K

32 GB

RTX 5090 · 32 GB tier

Magistral Small 2506

Q4_K_M · ~78 tok/s

17.8 GB · up to 40Kfaster

Qwen3.8 27B

Q4_K_M · ~67 tok/s

17.9 GB · up to 256K

48 GB

RTX A6000 · 48 GB tier

Magistral Small 2506

Q4_K_M · ~34 tok/s

17.8 GB · up to 40Kfaster

Qwen3.8 27B

Q4_K_M · ~29 tok/s

17.9 GB · up to 256K

Mac 32 GB

Mac Mini M4 · Macs

Magistral Small 2506

Q4_K_M · ~6 tok/s

17.8 GB · up to 40Kfaster

Qwen3.8 27B

Q4_K_M · ~5 tok/s

17.9 GB · up to 128K

Mac 64 GB

Mac Studio M4 Max · Macs

Magistral Small 2506

Q4_K_M · ~26 tok/s

17.8 GB · up to 40Kfaster

Qwen3.8 27B

Q4_K_M · ~22 tok/s

17.9 GB · up to 256K

Mac 128 GB

Mac Studio M4 Max · Macs

Magistral Small 2506

Q4_K_M · ~26 tok/s

17.8 GB · up to 40Kfaster

Qwen3.8 27B

Q4_K_M · ~22 tok/s

17.9 GB · up to 256K

Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).

llmrun Score

with its 90% range · bars share one scale

Magistral Small 2506 Qwen3.8 27B

Overall

Magistral Small 2506

88.784–93

Qwen3.8 27B

121.5119–124

Coding

Magistral Small 2506

no data

Qwen3.8 27B

112.1105–119

Math

Magistral Small 2506

no data

Qwen3.8 27B

117.5111–125

Science

Magistral Small 2506

no data

Qwen3.8 27B

112.093–131

Reasoning

Magistral Small 2506

88.272–104

Qwen3.8 27B

124.1121–127

Magistral Small 2506 · #65 of 92 open models · #19 of 34 that fit 24 GB

Qwen3.8 27B · #10 of 92 open models · #1 of 34 that fit 24 GB

Benchmarks

2 shared boards: Magistral Small 2506 leads 0, Qwen3.8 27B leads 2 · Benchmarks imported Oct 2026

BoardMagistral Small 2506Qwen3.8 27B

ARC-AGI-2

Magistral Small 2506

0.0%#18/18 open

IndependentEpoch AI · Jun 2025

Qwen3.8 27B

42.4%#5/18 open

IndependentEpoch AI · Aug 2026

ARC-AGI

Magistral Small 2506

5.0%#16/18 open

IndependentEpoch AI · Jun 2025

Qwen3.8 27B

87.5%#5/18 open

IndependentEpoch AI · Aug 2026

14 more boards (other shared and single-model boards)

Only Magistral Small 2506 (2)

GPQA Diamond56.1%
AIME 2024/202530.0%

Only Qwen3.8 27B (12)

VRAM by quantization

Memory to load the model and 2K context; below it, the total at 32K and 128K context. Lighter side in bold.

QuantMagistral Small 2506Qwen3.8 27B

Q2_K

Magistral Small 2506

10.7 GB8.9 GB file

32K 17.0 GB · 128K —

Qwen3.8 27B

12.6 GB10.8 GB file

32K 13.9 GB · 128K 19.3 GB

Q3_K_S

Magistral Small 2506

11.0 GB10.4 GB file

32K 17.3 GB · 128K —

Qwen3.8 27B

12.9 GB12.7 GB file

32K 14.2 GB · 128K 19.6 GB

Q3_K_M

Magistral Small 2506

12.2 GB11.5 GB file

32K 18.5 GB · 128K —

Qwen3.8 27B

14.3 GB13.4 GB file

32K 15.6 GB · 128K 21.0 GB

Q4_0

Magistral Small 2506

12.5 GB13.5 GB file

32K 18.8 GB · 128K —

Qwen3.8 27B

14.6 GB~13.9 GB file

32K 16.0 GB · 128K 21.4 GB

Q4_K_M

Magistral Small 2506

14.9 GB14.3 GB file

32K 21.2 GB · 128K —

Qwen3.8 27B

17.4 GB16.5 GB file

32K 18.8 GB · 128K 24.1 GB

Q5_K_M

Magistral Small 2506

17.5 GB16.8 GB file

32K 23.8 GB · 128K —

Qwen3.8 27B

20.5 GB19.8 GB file

32K 21.9 GB · 128K 27.3 GB

Q6_K

Magistral Small 2506

20.2 GB19.3 GB file

32K 26.5 GB · 128K —

Qwen3.8 27B

23.7 GB22.0 GB file

32K 25.0 GB · 128K 30.4 GB

Q8_0

Magistral Small 2506

24.3 GB25.1 GB file

32K 30.6 GB · 128K —

Qwen3.8 27B

28.5 GB29.0 GB file

32K 29.9 GB · 128K 35.2 GB

18 more quants

IQ2_XXS

Magistral Small 2506

7.2 GB6.5 GB file

32K 13.5 GB · 128K —

Qwen3.8 27B

8.4 GB7.3 GB file

32K 9.7 GB · 128K 15.1 GB

IQ2_XS

Magistral Small 2506

7.8 GB7.2 GB file

32K 14.1 GB · 128K —

Qwen3.8 27B

9.1 GB9.1 GB file

32K 10.4 GB · 128K 15.8 GB

IQ2_S

Magistral Small 2506

8.1 GB7.5 GB file

32K 14.4 GB · 128K —

Qwen3.8 27B

9.4 GB8.4 GB file

32K 10.8 GB · 128K 16.1 GB

IQ2_M

Magistral Small 2506

8.7 GB8.1 GB file

32K 15.0 GB · 128K —

Qwen3.8 27B

10.1 GB10.5 GB file

32K 11.5 GB · 128K 16.8 GB

IQ3_XXS

Magistral Small 2506

9.9 GB9.3 GB file

32K 16.1 GB · 128K —

Qwen3.8 27B

11.5 GB10.9 GB file

32K 12.9 GB · 128K 18.2 GB

Q2_K_S

Magistral Small 2506

10.2 GB8.3 GB file

32K 16.4 GB · 128K —

Qwen3.8 27B

not listed

IQ3_XS

Magistral Small 2506

10.4 GB9.9 GB file

32K 16.7 GB · 128K —

Qwen3.8 27B

12.2 GB10.4 GB file

32K 13.6 GB · 128K 18.9 GB

IQ3_S

Magistral Small 2506

10.7 GB10.4 GB file

32K 17.0 GB · 128K —

Qwen3.8 27B

12.6 GB12.0 GB file

32K 13.9 GB · 128K 19.3 GB

IQ3_M

Magistral Small 2506

11.3 GB10.7 GB file

32K 17.6 GB · 128K —

Qwen3.8 27B

13.3 GB14.9 GB file

32K 14.6 GB · 128K 20.0 GB

Q3_K_L

Magistral Small 2506

12.8 GB12.4 GB file

32K 19.1 GB · 128K —

Qwen3.8 27B

15.0 GB14.1 GB file

32K 16.3 GB · 128K 21.7 GB

IQ4_XS

Magistral Small 2506

13.4 GB12.8 GB file

32K 19.7 GB · 128K —

Qwen3.8 27B

15.7 GB14.3 GB file

32K 17.0 GB · 128K 22.4 GB

IQ4_NL

Magistral Small 2506

14.0 GB13.5 GB file

32K 20.3 GB · 128K —

Qwen3.8 27B

16.4 GB17.4 GB file

32K 17.7 GB · 128K 23.1 GB

Q4_1

Magistral Small 2506

14.0 GB14.9 GB file

32K 20.3 GB · 128K —

Qwen3.8 27B

16.4 GB17.5 GB file

32K 17.7 GB · 128K 23.1 GB

Q4_K_S

Magistral Small 2506

14.0 GB13.5 GB file

32K 20.3 GB · 128K —

Qwen3.8 27B

16.4 GB15.4 GB file

32K 17.7 GB · 128K 23.1 GB

Q4_K_L

Magistral Small 2506

15.2 GB14.8 GB file

32K 21.5 GB · 128K —

Qwen3.8 27B

17.8 GB18.8 GB file

32K 19.1 GB · 128K 24.5 GB

Q5_K_S

Magistral Small 2506

16.9 GB16.3 GB file

32K 23.2 GB · 128K —

Qwen3.8 27B

19.9 GB18.7 GB file

32K 21.2 GB · 128K 26.6 GB

Q5_K_L

Magistral Small 2506

17.8 GB17.2 GB file

32K 24.1 GB · 128K —

Qwen3.8 27B

20.9 GB21.5 GB file

32K 22.2 GB · 128K 27.6 GB

BF16

Magistral Small 2506

47.9 GB47.2 GB file

32K 54.2 GB · 128K —

Qwen3.8 27B

56.3 GB54.7 GB file

32K 57.7 GB · 128K 63.0 GB

How to run each

Magistral Small 2506

Ollama · Q4_K_M

$ollama run hf.co/bartowski/mistralai_Magistral-Small-2506-GGUF:Q4_K_M

All ways to run Magistral Small 2506 →

Qwen3.8 27B

Ollama · Q4_K_M

$ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:Q4_K_M

All ways to run Qwen3.8 27B →

Specifications

Specifications of Magistral Small 2506 and Qwen3.8 27B
Spec	Magistral Small 2506	Qwen3.8 27B
Publisher	Mistral AI	Alibaba
Parameters	23.6B	27.8B
Architecture	MistralForCausalLM	Qwen3_5ForConditionalGeneration
Context	40K tokens	256K tokens
Modalities	Chat	Vision, Function-calling
License	Apache 2.0	Apache 2.0
Released	Jun 2025	Aug 2026

Frequently Asked Questions

Which is better, Magistral Small 2506 or Qwen3.8 27B?

Qwen3.8 27B leads llmrun Score 121.5 to 88.7; both fit 24 GB at Q4_K_M. Choose Magistral Small 2506 for also does chat; choose Qwen3.8 27B for higher llmrun Score and stronger at reasoning.

Which needs less VRAM, Magistral Small 2506 or Qwen3.8 27B?

At Q4_K_M, Magistral Small 2506 requires 14.9 GB while Qwen3.8 27B needs 17.4 GB, making Magistral Small 2506 the easier model to run on your local setup.

Which is faster on RTX 4090?

Estimated decode speed on RTX 4090 is about 44 tok/s for Magistral Small 2506 (Q4_K_M) and 38 tok/s for Qwen3.8 27B (Q4_K_M), so Magistral Small 2506 is faster. Estimates, ±20%.

Can I run Magistral Small 2506 or Qwen3.8 27B on a 16 GB GPU?

Magistral Small 2506 fits at Q4_K_M; Qwen3.8 27B fits at Q3_K_L.

Which has a longer context window, Magistral Small 2506 or Qwen3.8 27B?

Magistral Small 2506 supports 40,960 tokens and Qwen3.8 27B supports 262,144 tokens.