llmrun

Phi 4 vs GPT OSS 20B

GPT OSS 20B leads llmrun Score 96.2 to 83.8; both fit 24 GB at Q4_K_M; GPT OSS 20B is 2.8× faster on RTX 4090 (~190 vs ~69 tok/s).

Model A · Microsoft

Phi 4
83.8llmrun Score

14.7B · 16K context · MIT · Dec 2024

ChatMathCode

Model B · OpenAI

GPT OSS 20B
96.2llmrun Score

20.9B · 4.2B active · 128K context · Apache 2.0 · Aug 2025

ChatFunctions

Which should you choose?

Choose Phi 4 if you want…

  • Lighter on memory

    9.5 vs 12.9 GB at Q4_K_M

  • Also does math, code

    GPT OSS 20B doesn't list it

Choose GPT OSS 20B if you want…

  • Higher llmrun Score

    96.2 vs 83.8

  • Ahead on the benchmarks

    5 of 5 shared boards

  • 2.8× faster on RTX 4090

    ~190 vs ~69 tok/s

  • Longer context window

    128K vs 16K tokens

  • Also does function-calling

    Phi 4 doesn't list it

  • Newer release

    Aug 2025 vs Dec 2024

On your hardware

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

8 GB

RTX 4060 · 8 GB tier

Phi 4

IQ2_XXS · ~37 tok/s

7.7 GB · up to 16K

GPT OSS 20B

IQ2_M · ~103 tok/s

7.6 GB · up to 32K

12 GB

RTX 3060 12GB · 12 GB tier

Phi 4

Q4_K_M · ~25 tok/stight

12.5 GB · up to 8K

GPT OSS 20B

IQ4_XS · ~89 tok/s

11.8 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 16K

GPT OSS 20B

Q4_K_M · ~98 tok/s

13.1 GB · up to 128Kfaster

24 GBreference

RTX 4090 · 24 GB tier

Phi 4

Q4_K_M · ~69 tok/s

12.5 GB · up to 16K

GPT OSS 20B

Q4_K_M · ~190 tok/s

13.1 GB · up to 128Kfaster

32 GB

RTX 5090 · 32 GB tier

Phi 4

Q4_K_M · ~122 tok/s

12.5 GB · up to 16K

GPT OSS 20B

Q4_K_M · ~282 tok/s

13.1 GB · up to 128Kfaster

48 GB

RTX A6000 · 48 GB tier

Phi 4

Q4_K_M · ~52 tok/s

12.5 GB · up to 16K

GPT OSS 20B

Q4_K_M · ~154 tok/s

13.1 GB · up to 128Kfaster

Mac 32 GB

Mac Mini M4 · Macs

Phi 4

Q4_K_M · ~9 tok/s

12.5 GB · up to 16K

GPT OSS 20B

Q4_K_M · ~28 tok/s

13.1 GB · up to 128Kfaster

Mac 64 GB

Mac Studio M4 Max · Macs

Phi 4

Q4_K_M · ~40 tok/s

12.5 GB · up to 16K

GPT OSS 20B

Q4_K_M · ~86 tok/s

13.1 GB · up to 128Kfaster

Mac 128 GB

Mac Studio M4 Max · Macs

Phi 4

Q4_K_M · ~40 tok/s

12.5 GB · up to 16K

GPT OSS 20B

Q4_K_M · ~86 tok/s

13.1 GB · up to 128Kfaster

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

Overall

Phi 4

83.879–89

GPT OSS 20B

96.293–99

Coding

Phi 4

no data

GPT OSS 20B

100.992–110

Science

Phi 4

no data

GPT OSS 20B

92.485–100

Reasoning

Phi 4

no data

GPT OSS 20B

90.882–99

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

GPT OSS 20B · #58 of 92 open models · #15 of 34 that fit 24 GB

Benchmarks

5 shared boards: Phi 4 leads 0, GPT OSS 20B leads 5 · Benchmarks imported Oct 2026

GPQA Diamond

Phi 4

56.1%#45/97 open

IndependentEpoch AI · Dec 2024

GPT OSS 20B

60.8%#41/97 open

IndependentEpoch AI · Aug 2025

LMArena Text

Phi 4

1256.3#127/186 open

ArenaLMArena · Oct 2026

GPT OSS 20B

1317.3#96/186 open

ArenaLMArena · Oct 2026

AIME 2024/2025

Phi 4

13.8%#56/85 open

IndependentEpoch AI · Dec 2024

GPT OSS 20B

65.3%#33/85 open

IndependentEpoch AI · Aug 2025

MMLU-Pro

Phi 4

70.4%#37/144 open

IndependentTIGER-Lab

GPT OSS 20B

73.6%#33/144 open

IndependentTIGER-Lab

Chess Puzzles

Phi 4

1.0%#42/67 open

IndependentEpoch AI · Dec 2024

GPT OSS 20B

4.0%#33/67 open

IndependentEpoch AI · Aug 2025

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

Only Phi 4 (3)

Only GPT OSS 20B (7)

VRAM by quantization

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

Q2_K

Phi 4

7.0 GB5.5 GB file

32K — · 128K —

GPT OSS 20B

9.3 GB11.5 GB file

32K 9.8 GB · 128K 11.5 GB

Q3_K_S

Phi 4

7.1 GB6.5 GB file

32K — · 128K —

GPT OSS 20B

9.5 GB11.5 GB file

32K 10.0 GB · 128K 11.7 GB

Q3_K_M

Phi 4

7.9 GB7.4 GB file

32K — · 128K —

GPT OSS 20B

10.6 GB11.5 GB file

32K 11.1 GB · 128K 12.8 GB

Q4_0

Phi 4

8.1 GB8.4 GB file

32K — · 128K —

GPT OSS 20B

10.8 GB11.5 GB file

32K 11.3 GB · 128K 13.0 GB

Q4_K_M

Phi 4

9.5 GB9.1 GB file

32K — · 128K —

GPT OSS 20B

12.9 GB11.6 GB file

32K 13.4 GB · 128K 15.1 GB

Q5_K_M

Phi 4

11.2 GB10.6 GB file

32K — · 128K —

GPT OSS 20B

15.3 GB11.7 GB file

32K 15.8 GB · 128K 17.5 GB

Q6_K

Phi 4

12.8 GB12.0 GB file

32K — · 128K —

GPT OSS 20B

17.6 GB12.0 GB file

32K 18.1 GB · 128K 19.8 GB

Q8_0

Phi 4

15.4 GB15.6 GB file

32K — · 128K —

GPT OSS 20B

21.3 GB~20.9 GB file

32K 21.8 GB · 128K 23.5 GB

19 more quants

IQ2_XXS

Phi 4

4.8 GB4.1 GB file

32K — · 128K —

GPT OSS 20B

6.1 GB~5.8 GB file

32K 6.6 GB · 128K 8.3 GB

IQ2_XS

Phi 4

5.1 GB4.5 GB file

32K — · 128K —

GPT OSS 20B

6.7 GB~6.3 GB file

32K 7.2 GB · 128K 8.9 GB

IQ2_S

Phi 4

5.3 GB4.7 GB file

32K — · 128K —

GPT OSS 20B

6.9 GB~6.5 GB file

32K 7.4 GB · 128K 9.1 GB

IQ2_M

Phi 4

5.7 GB5.1 GB file

32K — · 128K —

GPT OSS 20B

7.4 GB~7.1 GB file

32K 7.9 GB · 128K 9.6 GB

IQ3_XXS

Phi 4

6.4 GB5.8 GB file

32K — · 128K —

GPT OSS 20B

8.5 GB11.6 GB file

32K 9.0 GB · 128K 10.7 GB

Q2_K_S

Phi 4

6.6 GB5.2 GB file

32K — · 128K —

GPT OSS 20B

8.7 GB12.1 GB file

32K 9.2 GB · 128K 10.9 GB

IQ3_XS

Phi 4

6.8 GB6.2 GB file

32K — · 128K —

GPT OSS 20B

9.0 GB11.6 GB file

32K 9.5 GB · 128K 11.2 GB

IQ3_S

Phi 4

7.0 GB6.5 GB file

32K — · 128K —

GPT OSS 20B

9.3 GB12.1 GB file

32K 9.8 GB · 128K 11.5 GB

IQ3_M

Phi 4

7.3 GB6.9 GB file

32K — · 128K —

GPT OSS 20B

9.8 GB11.6 GB file

32K 10.3 GB · 128K 12.0 GB

Q3_K_L

Phi 4

8.2 GB7.9 GB file

32K — · 128K —

GPT OSS 20B

11.1 GB11.5 GB file

32K 11.6 GB · 128K 13.3 GB

IQ4_XS

Phi 4

8.6 GB7.9 GB file

32K — · 128K —

GPT OSS 20B

11.6 GB11.6 GB file

32K 12.1 GB · 128K 13.8 GB

IQ4_NL

Phi 4

9.0 GB8.4 GB file

32K — · 128K —

GPT OSS 20B

12.1 GB11.6 GB file

32K 12.6 GB · 128K 14.3 GB

Q4_1

Phi 4

9.0 GB9.3 GB file

32K — · 128K —

GPT OSS 20B

12.1 GB11.6 GB file

32K 12.6 GB · 128K 14.3 GB

Q4_K_S

Phi 4

9.0 GB8.4 GB file

32K — · 128K —

GPT OSS 20B

12.1 GB11.6 GB file

32K 12.6 GB · 128K 14.3 GB

Q4_K_L

Phi 4

9.7 GB9.4 GB file

32K — · 128K —

GPT OSS 20B

13.2 GB11.9 GB file

32K 13.7 GB · 128K 15.4 GB

Q5_0

Phi 4

9.9 GB10.2 GB file

32K — · 128K —

GPT OSS 20B

not listed

Q5_K_S

Phi 4

10.8 GB10.2 GB file

32K — · 128K —

GPT OSS 20B

14.8 GB11.7 GB file

32K 15.3 GB · 128K 17.0 GB

Q5_K_L

Phi 4

11.4 GB10.9 GB file

32K — · 128K —

GPT OSS 20B

15.5 GB11.9 GB file

32K 16.0 GB · 128K 17.7 GB

BF16

Phi 4

30.0 GB~29.3 GB file

32K — · 128K —

GPT OSS 20B

42.2 GB~41.8 GB file

32K 42.7 GB · 128K 44.4 GB

How to run each

Specifications

Specifications of Phi 4 and GPT OSS 20B
SpecPhi 4GPT OSS 20B
PublisherMicrosoftOpenAI
Parameters14.7B20.9B
Active parameters—4.2B per token
ArchitecturePhi3ForCausalLMGptOssForCausalLM
Context16K tokens128K tokens
ModalitiesChat, Math, CodeChat, Function-calling
LicenseMITApache 2.0
ReleasedDec 2024Aug 2025

Frequently Asked Questions

Which is better, Phi 4 or GPT OSS 20B?

GPT OSS 20B leads llmrun Score 96.2 to 83.8; both fit 24 GB at Q4_K_M; GPT OSS 20B is 2.8× faster on RTX 4090 (~190 vs ~69 tok/s). Choose Phi 4 for lighter on memory and also does math, code; choose GPT OSS 20B for higher llmrun Score and ahead on the benchmarks.

Which needs less VRAM, Phi 4 or GPT OSS 20B?

At Q4_K_M, Phi 4 requires 9.5 GB while GPT OSS 20B needs 12.9 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 190 tok/s for GPT OSS 20B (Q4_K_M), so GPT OSS 20B is faster. Estimates, ±20%.

Can I run Phi 4 or GPT OSS 20B on a 16 GB GPU?

Phi 4 fits at Q4_K_M; GPT OSS 20B fits at Q4_K_M.

Which has a longer context window, Phi 4 or GPT OSS 20B?

Phi 4 supports 16,384 tokens and GPT OSS 20B supports 131,072 tokens.

Phi 4 vs GPT OSS 20B — VRAM & GPU Comparison | llmrun