Phi 4 vs Magistral Small 2506

Magistral Small 2506 leads llmrun Score 88.7 to 83.8 (ranges overlap); both fit 24 GB at Q4_K_M; Phi 4 is 1.6× faster on RTX 4090 (~69 vs ~44 tok/s).

Model A · Microsoft

Phi 4

83.8llmrun Score

14.7B · 16K context · MIT · Dec 2024

ChatMathCode

Model B · Mistral AI

Magistral Small 2506

88.7llmrun Score

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

Chat

Which should you choose?

Choose Phi 4 if you want…

Fits a 12 GB GPU
Q4_K_M · 12.5 GB; Magistral Small 2506 needs 10.1 GB at its smallest quant
1.6× faster on RTX 4090
~69 vs ~44 tok/s
Lighter on memory
9.5 vs 14.9 GB at Q4_K_M
Also does math, code
Magistral Small 2506 doesn't list it

Choose Magistral Small 2506 if you want…

Higher llmrun Score
88.7 vs 83.8, ranges overlap
Longer context window
40K vs 16K tokens
Newer release
Jun 2025 vs Dec 2024

On your hardware

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

MemoryPhi 4Magistral Small 2506

8 GB

RTX 4060 · 8 GB tier

Phi 4

IQ2_XXS · ~37 tok/s

7.7 GB · up to 16Kfits

Magistral Small 2506

Doesn't fit

smallest quant needs 10.1 GB

12 GB

RTX 3060 12GB · 12 GB tier

Phi 4

Q4_K_M · ~25 tok/stight

12.5 GB · up to 8K

Magistral Small 2506

IQ2_M · ~27 tok/s

11.6 GB · up to 16Kmore headroom

16 GB

RTX 5060 Ti 16GB · 16 GB tier

Phi 4

Q4_K_M · ~31 tok/s

12.5 GB · up to 16Kmore headroom

Magistral Small 2506

Q4_K_M · ~20 tok/stight

17.8 GB · up to 4K

24 GBreference

RTX 4090 · 24 GB tier

Phi 4

Q4_K_M · ~69 tok/s

12.5 GB · up to 16Kfaster

Magistral Small 2506

Q4_K_M · ~44 tok/s

17.8 GB · up to 40K

32 GB

RTX 5090 · 32 GB tier

Phi 4

Q4_K_M · ~122 tok/s

12.5 GB · up to 16Kfaster

Magistral Small 2506

Q4_K_M · ~78 tok/s

17.8 GB · up to 40K

48 GB

RTX A6000 · 48 GB tier

Phi 4

Q4_K_M · ~52 tok/s

12.5 GB · up to 16Kfaster

Magistral Small 2506

Q4_K_M · ~34 tok/s

17.8 GB · up to 40K

Mac 32 GB

Mac Mini M4 · Macs

Phi 4

Q4_K_M · ~9 tok/s

12.5 GB · up to 16Kfaster

Magistral Small 2506

Q4_K_M · ~6 tok/s

17.8 GB · up to 40K

Mac 64 GB

Mac Studio M4 Max · Macs

Phi 4

Q4_K_M · ~40 tok/s

12.5 GB · up to 16Kfaster

Magistral Small 2506

Q4_K_M · ~26 tok/s

17.8 GB · up to 40K

Mac 128 GB

Mac Studio M4 Max · Macs

Phi 4

Q4_K_M · ~40 tok/s

12.5 GB · up to 16Kfaster

Magistral Small 2506

Q4_K_M · ~26 tok/s

17.8 GB · up to 40K

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

Phi 4 Magistral Small 2506

Overall

Phi 4

83.879–89

Magistral Small 2506

88.784–93

Reasoning

Phi 4

no data

Magistral Small 2506

88.272–104

Phi 4 · #70 of 92 open models · #22 of 34 that fit 24 GB

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

Benchmarks

2 shared boards: Phi 4 leads 0, Magistral Small 2506 leads 1 · Benchmarks imported Oct 2026

BoardPhi 4Magistral Small 2506

GPQA Diamond

Phi 4

56.1%#45/97 open

IndependentEpoch AI · Dec 2024

Magistral Small 2506

56.1%#46/97 open

IndependentEpoch AI · Jun 2025

AIME 2024/2025

Phi 4

13.8%#56/85 open

IndependentEpoch AI · Dec 2024

Magistral Small 2506

30.0%#49/85 open

IndependentEpoch AI · Jun 2025

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

Only Phi 4 (6)

Only Magistral Small 2506 (2)

ARC-AGI-20.0%
ARC-AGI5.0%

VRAM by quantization

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

QuantPhi 4Magistral Small 2506

Q2_K

Phi 4

7.0 GB5.5 GB file

32K — · 128K —

Magistral Small 2506

10.7 GB8.9 GB file

32K 17.0 GB · 128K —

Q3_K_S

Phi 4

7.1 GB6.5 GB file

32K — · 128K —

Magistral Small 2506

11.0 GB10.4 GB file

32K 17.3 GB · 128K —

Q3_K_M

Phi 4

7.9 GB7.4 GB file

32K — · 128K —

Magistral Small 2506

12.2 GB11.5 GB file

32K 18.5 GB · 128K —

Q4_0

Phi 4

8.1 GB8.4 GB file

32K — · 128K —

Magistral Small 2506

12.5 GB13.5 GB file

32K 18.8 GB · 128K —

Q4_K_M

Phi 4

9.5 GB9.1 GB file

32K — · 128K —

Magistral Small 2506

14.9 GB14.3 GB file

32K 21.2 GB · 128K —

Q5_K_M

Phi 4

11.2 GB10.6 GB file

32K — · 128K —

Magistral Small 2506

17.5 GB16.8 GB file

32K 23.8 GB · 128K —

Q6_K

Phi 4

12.8 GB12.0 GB file

32K — · 128K —

Magistral Small 2506

20.2 GB19.3 GB file

32K 26.5 GB · 128K —

Q8_0

Phi 4

15.4 GB15.6 GB file

32K — · 128K —

Magistral Small 2506

24.3 GB25.1 GB file

32K 30.6 GB · 128K —

19 more quants

IQ2_XXS

Phi 4

4.8 GB4.1 GB file

32K — · 128K —

Magistral Small 2506

7.2 GB6.5 GB file

32K 13.5 GB · 128K —

IQ2_XS

Phi 4

5.1 GB4.5 GB file

32K — · 128K —

Magistral Small 2506

7.8 GB7.2 GB file

32K 14.1 GB · 128K —

IQ2_S

Phi 4

5.3 GB4.7 GB file

32K — · 128K —

Magistral Small 2506

8.1 GB7.5 GB file

32K 14.4 GB · 128K —

IQ2_M

Phi 4

5.7 GB5.1 GB file

32K — · 128K —

Magistral Small 2506

8.7 GB8.1 GB file

32K 15.0 GB · 128K —

IQ3_XXS

Phi 4

6.4 GB5.8 GB file

32K — · 128K —

Magistral Small 2506

9.9 GB9.3 GB file

32K 16.1 GB · 128K —

Q2_K_S

Phi 4

6.6 GB5.2 GB file

32K — · 128K —

Magistral Small 2506

10.2 GB8.3 GB file

32K 16.4 GB · 128K —

IQ3_XS

Phi 4

6.8 GB6.2 GB file

32K — · 128K —

Magistral Small 2506

10.4 GB9.9 GB file

32K 16.7 GB · 128K —

IQ3_S

Phi 4

7.0 GB6.5 GB file

32K — · 128K —

Magistral Small 2506

10.7 GB10.4 GB file

32K 17.0 GB · 128K —

IQ3_M

Phi 4

7.3 GB6.9 GB file

32K — · 128K —

Magistral Small 2506

11.3 GB10.7 GB file

32K 17.6 GB · 128K —

Q3_K_L

Phi 4

8.2 GB7.9 GB file

32K — · 128K —

Magistral Small 2506

12.8 GB12.4 GB file

32K 19.1 GB · 128K —

IQ4_XS

Phi 4

8.6 GB7.9 GB file

32K — · 128K —

Magistral Small 2506

13.4 GB12.8 GB file

32K 19.7 GB · 128K —

IQ4_NL

Phi 4

9.0 GB8.4 GB file

32K — · 128K —

Magistral Small 2506

14.0 GB13.5 GB file

32K 20.3 GB · 128K —

Q4_1

Phi 4

9.0 GB9.3 GB file

32K — · 128K —

Magistral Small 2506

14.0 GB14.9 GB file

32K 20.3 GB · 128K —

Q4_K_S

Phi 4

9.0 GB8.4 GB file

32K — · 128K —

Magistral Small 2506

14.0 GB13.5 GB file

32K 20.3 GB · 128K —

Q4_K_L

Phi 4

9.7 GB9.4 GB file

32K — · 128K —

Magistral Small 2506

15.2 GB14.8 GB file

32K 21.5 GB · 128K —

Q5_0

Phi 4

9.9 GB10.2 GB file

32K — · 128K —

Magistral Small 2506

not listed

Q5_K_S

Phi 4

10.8 GB10.2 GB file

32K — · 128K —

Magistral Small 2506

16.9 GB16.3 GB file

32K 23.2 GB · 128K —

Q5_K_L

Phi 4

11.4 GB10.9 GB file

32K — · 128K —

Magistral Small 2506

17.8 GB17.2 GB file

32K 24.1 GB · 128K —

BF16

Phi 4

30.0 GB~29.3 GB file

32K — · 128K —

Magistral Small 2506

47.9 GB47.2 GB file

32K 54.2 GB · 128K —

How to run each

Phi 4

Ollama · Q4_K_M

$ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q4_K_M

All ways to run Phi 4 →

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 →

Specifications

Specifications of Phi 4 and Magistral Small 2506
Spec	Phi 4	Magistral Small 2506
Publisher	Microsoft	Mistral AI
Parameters	14.7B	23.6B
Architecture	Phi3ForCausalLM	MistralForCausalLM
Context	16K tokens	40K tokens
Modalities	Chat, Math, Code	Chat
License	MIT	Apache 2.0
Released	Dec 2024	Jun 2025

Frequently Asked Questions

Which is better, Phi 4 or Magistral Small 2506?

Magistral Small 2506 leads llmrun Score 88.7 to 83.8 (ranges overlap); both fit 24 GB at Q4_K_M; Phi 4 is 1.6× faster on RTX 4090 (~69 vs ~44 tok/s). Choose Phi 4 for fits a 12 GB GPU and 1.6× faster on RTX 4090; choose Magistral Small 2506 for higher llmrun Score and longer context window.

Which needs less VRAM, Phi 4 or Magistral Small 2506?

At Q4_K_M, Phi 4 requires 9.5 GB while Magistral Small 2506 needs 14.9 GB, making Phi 4 the easier model to run on your local setup.

Which is faster on RTX 4090?

Estimated decode speed on RTX 4090 is about 69 tok/s for Phi 4 (Q4_K_M) and 44 tok/s for Magistral Small 2506 (Q4_K_M), so Phi 4 is faster. Estimates, ±20%.

Can I run Phi 4 or Magistral Small 2506 on a 16 GB GPU?

Phi 4 fits at Q4_K_M; Magistral Small 2506 fits at Q4_K_M.

Which has a longer context window, Phi 4 or Magistral Small 2506?

Phi 4 supports 16,384 tokens and Magistral Small 2506 supports 40,960 tokens.