Phi 4 vs Qwen3.6 35B A3B
Qwen3.6 35B A3B leads llmrun Score 106.8 to 83.9; both fit 24 GB at Q4_K_M; Qwen3.6 35B A3B is 2.2× faster on RTX 4090 (~151 vs ~69 tok/s).
Model B · Alibaba
Qwen3.6 35B A3B36.0B · 4.7B active · 256K context · Apache 2.0 · Apr 2026
Choose Phi 4 if you want…
Fits a 12 GB GPU
Q4_K_M · 12.5 GB; Qwen3.6 35B A3B needs 10.4 GB at its smallest quant
Lighter on memory
9.5 vs 22.0 GB at Q4_K_M
Also does chat, math, code
Qwen3.6 35B A3B doesn't list it
Choose Qwen3.6 35B A3B if you want…
Higher llmrun Score
106.8 vs 83.9
Stronger at science
110.1 vs 88.7
Ahead on the benchmarks
3 of 3 shared boards
2.2× faster on RTX 4090
~151 vs ~69 tok/s
Longer context window
256K vs 16K tokens
Also does vision, function-calling
Phi 4 doesn't list it
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 16Kfits
Qwen3.6 35B A3B
Doesn't fit
smallest quant needs 10.4 GB
12 GB
RTX 3060 12GB · 12 GB tier
Phi 4
Q4_K_M · ~25 tok/stight
12.5 GB · up to 8K
Qwen3.6 35B A3B
IQ2_S · ~114 tok/s
11.7 GB · up to 32Kmore 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
Qwen3.6 35B A3B
Q2_K · ~108 tok/s
15.7 GB · up to 32K
24 GBreference
RTX 4090 · 24 GB tier
Phi 4
Q4_K_M · ~69 tok/s
12.5 GB · up to 16K
Qwen3.6 35B A3B
Q4_K_M · ~151 tok/s
22 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
Qwen3.6 35B A3B
Q4_K_M · ~213 tok/s
22 GB · up to 256Kfaster
48 GB
RTX A6000 · 48 GB tier
Phi 4
Q4_K_M · ~52 tok/s
12.5 GB · up to 16K
Qwen3.6 35B A3B
Q4_K_M · ~125 tok/s
22 GB · up to 256Kfaster
Mac 32 GB
Mac Mini M4 · Macs
Phi 4
Q4_K_M · ~9 tok/s
12.5 GB · up to 16K
Qwen3.6 35B A3B
Q4_K_M · ~24 tok/s
22 GB · up to 256Kfaster
Mac 64 GB
Mac Studio M4 Max · Macs
Phi 4
Q4_K_M · ~40 tok/s
12.5 GB · up to 16K
Qwen3.6 35B A3B
Q4_K_M · ~64 tok/s
22 GB · up to 256Kfaster
Mac 128 GB
Mac Studio M4 Max · Macs
Phi 4
Q4_K_M · ~40 tok/s
12.5 GB · up to 16K
Qwen3.6 35B A3B
Q4_K_M · ~64 tok/s
22 GB · up to 256Kfaster
Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).
Overall
Phi 4
Qwen3.6 35B A3B
Coding
Phi 4
no data
Qwen3.6 35B A3B
Math
Phi 4
no data
Qwen3.6 35B A3B
Science
Phi 4
Qwen3.6 35B A3B
Reasoning
Phi 4
no data
Qwen3.6 35B A3B
Phi 4 · #81 of 106 open models · #23 of 36 that fit 24 GB
Qwen3.6 35B A3B · #41 of 106 open models · #3 of 36 that fit 24 GB
Phi 4
56.1%#48/100 open
IndependentEpoch AI · Dec 2024
Qwen3.6 35B A3B
84.9%#23/100 open
IndependentEpoch AI · Apr 2026
Phi 4
13.8%#59/88 open
IndependentEpoch AI · Dec 2024
Qwen3.6 35B A3B
86.7%#22/88 open
IndependentEpoch AI · Apr 2026
Phi 4
0.0%#69/84 open
Lab-reportedEpoch AI · Dec 2024
Qwen3.6 35B A3B
0.3%#51/84 open
Lab-reportedEpoch AI · Apr 2026
13 more boards (other shared and single-model boards)
Only Phi 4 (5)
- LMArena Text1256.3
- MMLU-Pro70.4%
- MATH Level 564.9%
- MMLU84.8%
- BALROG11.6%
Only Qwen3.6 35B A3B (8)
- FrontierMath20.3%
- Terminal-Bench24.6%
- SciCode36.6%
- WeirdML34.5%
- DTBench73.9%
- LMCA29.7%
- Mystery Game Puzzles22.0%
- Surface Evolver Bench44.4%
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 —
Qwen3.6 35B A3B
15.7 GB13.5 GB file
32K 15.9 GB · 128K 16.9 GB
Q3_K_S
Phi 4
7.1 GB6.5 GB file
32K — · 128K —
Qwen3.6 35B A3B
16.1 GB15.4 GB file
32K 16.4 GB · 128K 17.4 GB
Q3_K_M
Phi 4
7.9 GB7.4 GB file
32K — · 128K —
Qwen3.6 35B A3B
17.9 GB16.6 GB file
32K 18.2 GB · 128K 19.2 GB
Q4_0
Phi 4
8.1 GB8.4 GB file
32K — · 128K —
Qwen3.6 35B A3B
18.4 GB19.8 GB file
32K 18.6 GB · 128K 19.6 GB
Q4_K_M
Phi 4
9.5 GB9.1 GB file
32K — · 128K —
Qwen3.6 35B A3B
22.0 GB22.1 GB file
32K 22.2 GB · 128K 23.2 GB
Q5_K_M
Phi 4
11.2 GB10.6 GB file
32K — · 128K —
Qwen3.6 35B A3B
26.0 GB26.5 GB file
32K 26.3 GB · 128K 27.3 GB
Q6_K
Phi 4
12.8 GB12.0 GB file
32K — · 128K —
Qwen3.6 35B A3B
30.0 GB29.3 GB file
32K 30.3 GB · 128K 31.3 GB
Q8_0
Phi 4
15.4 GB15.6 GB file
32K — · 128K —
Qwen3.6 35B A3B
36.3 GB36.9 GB file
32K 36.6 GB · 128K 37.6 GB
20 more quants
IQ2_XXS
Phi 4
4.8 GB4.1 GB file
32K — · 128K —
Qwen3.6 35B A3B
10.3 GB10.8 GB file
32K 10.5 GB · 128K 11.5 GB
IQ2_XS
Phi 4
5.1 GB4.5 GB file
32K — · 128K —
Qwen3.6 35B A3B
11.2 GB11.7 GB file
32K 11.4 GB · 128K 12.4 GB
IQ2_S
Phi 4
5.3 GB4.7 GB file
32K — · 128K —
Qwen3.6 35B A3B
11.6 GB11.9 GB file
32K 11.9 GB · 128K 12.9 GB
IQ2_M
Phi 4
5.7 GB5.1 GB file
32K — · 128K —
Qwen3.6 35B A3B
12.5 GB11.5 GB file
32K 12.8 GB · 128K 13.8 GB
IQ3_XXS
Phi 4
6.4 GB5.8 GB file
32K — · 128K —
Qwen3.6 35B A3B
14.3 GB13.2 GB file
32K 14.6 GB · 128K 15.6 GB
Q2_K_S
Phi 4
6.6 GB5.2 GB file
32K — · 128K —
Qwen3.6 35B A3B
14.8 GB12.2 GB file
32K 15.0 GB · 128K 16.0 GB
IQ3_XS
Phi 4
6.8 GB6.2 GB file
32K — · 128K —
Qwen3.6 35B A3B
15.2 GB17.1 GB file
32K 15.5 GB · 128K 16.5 GB
IQ3_S
Phi 4
7.0 GB6.5 GB file
32K — · 128K —
Qwen3.6 35B A3B
15.7 GB13.7 GB file
32K 15.9 GB · 128K 16.9 GB
IQ3_M
Phi 4
7.3 GB6.9 GB file
32K — · 128K —
Qwen3.6 35B A3B
16.6 GB17.8 GB file
32K 16.8 GB · 128K 17.8 GB
Q3_K_L
Phi 4
8.2 GB7.9 GB file
32K — · 128K —
Qwen3.6 35B A3B
18.8 GB17.8 GB file
32K 19.1 GB · 128K 20.1 GB
IQ4_XS
Phi 4
8.6 GB7.9 GB file
32K — · 128K —
Qwen3.6 35B A3B
19.7 GB17.7 GB file
32K 20.0 GB · 128K 21.0 GB
IQ4_NL
Phi 4
9.0 GB8.4 GB file
32K — · 128K —
Qwen3.6 35B A3B
20.6 GB18.0 GB file
32K 20.9 GB · 128K 21.9 GB
Q4_1
Phi 4
9.0 GB9.3 GB file
32K — · 128K —
Qwen3.6 35B A3B
20.6 GB22.9 GB file
32K 20.9 GB · 128K 21.9 GB
Q4_K_S
Phi 4
9.0 GB8.4 GB file
32K — · 128K —
Qwen3.6 35B A3B
20.6 GB20.9 GB file
32K 20.9 GB · 128K 21.9 GB
Q4_K_L
Phi 4
9.7 GB9.4 GB file
32K — · 128K —
Qwen3.6 35B A3B
22.4 GB22.7 GB file
32K 22.7 GB · 128K 23.7 GB
Q5_0
Phi 4
9.9 GB10.2 GB file
32K — · 128K —
Qwen3.6 35B A3B
22.9 GB24.1 GB file
32K 23.1 GB · 128K 24.1 GB
Q5_1
Phi 4
not listed
Qwen3.6 35B A3B
25.1 GB26.1 GB file
32K 25.4 GB · 128K 26.4 GB
Q5_K_S
Phi 4
10.8 GB10.2 GB file
32K — · 128K —
Qwen3.6 35B A3B
25.1 GB24.9 GB file
32K 25.4 GB · 128K 26.4 GB
Q5_K_L
Phi 4
11.4 GB10.9 GB file
32K — · 128K —
Qwen3.6 35B A3B
26.5 GB26.2 GB file
32K 26.7 GB · 128K 27.7 GB
BF16
Phi 4
30.0 GB~29.3 GB file
32K — · 128K —
Qwen3.6 35B A3B
72.3 GB69.4 GB file
32K 72.5 GB · 128K 73.6 GB
Phi 4
Ollama · Q4_K_M
ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q4_K_MQwen3.6 35B A3B
Ollama · Q4_K_M
ollama run hf.co/unsloth/Qwen3.6-35B-A3B-GGUF:Q4_K_M| Spec | Phi 4 | Qwen3.6 35B A3B |
|---|---|---|
| Publisher | Microsoft | Alibaba |
| Parameters | 14.7B | 36.0B |
| Active parameters | — | 4.7B per token |
| Architecture | Phi3ForCausalLM | Qwen3_5MoeForConditionalGeneration |
| Context | 16K tokens | 256K tokens |
| Modalities | Chat, Math, Code | Vision, Function-calling |
| License | MIT | Apache 2.0 |
| Released | Dec 2024 | Apr 2026 |
Frequently Asked Questions
Which is better, Phi 4 or Qwen3.6 35B A3B?
Qwen3.6 35B A3B leads llmrun Score 106.8 to 83.9; both fit 24 GB at Q4_K_M; Qwen3.6 35B A3B is 2.2× faster on RTX 4090 (~151 vs ~69 tok/s). Choose Phi 4 for fits a 12 GB GPU and lighter on memory; choose Qwen3.6 35B A3B for higher llmrun Score and stronger at science.
Which needs less VRAM, Phi 4 or Qwen3.6 35B A3B?
At Q4_K_M, Phi 4 requires 9.5 GB while Qwen3.6 35B A3B needs 22.0 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 151 tok/s for Qwen3.6 35B A3B (Q4_K_M), so Qwen3.6 35B A3B is faster. Estimates, ±20%.
Can I run Phi 4 or Qwen3.6 35B A3B on a 16 GB GPU?
Phi 4 fits at Q4_K_M; Qwen3.6 35B A3B fits at Q2_K.
Which has a longer context window, Phi 4 or Qwen3.6 35B A3B?
Phi 4 supports 16,384 tokens and Qwen3.6 35B A3B supports 262,144 tokens.