CanItRun Logocanitrun.

Qwen 3.5 35B-A3B (MoE) vs Qwen 3.5 122B-A10B (MoE)

Side-by-side VRAM requirements, benchmark scores, and GPU compatibility for local AI inference.

Quick verdict

Qwen 3.5 35B-A3B (MoE) is more hardware-efficient: it needs 24.1 GB at its Q4_K_M build vs 83.4 GB for Qwen 3.5 122B-A10B (MoE)'s Q4_K_M, fitting on 84 GPUs natively.

VRAM at each quantization (8k context)

FP32
Qwen 3.5 35B-A3B (MoE)
157.0 GB
Qwen 3.5 122B-A10B (MoE)
546.8 GB
BF16
Qwen 3.5 35B-A3B (MoE)
78.6 GB
Qwen 3.5 122B-A10B (MoE)
273.5 GB
FP16
Qwen 3.5 35B-A3B (MoE)
78.6 GB
Qwen 3.5 122B-A10B (MoE)
273.5 GB
Q8_0
Qwen 3.5 35B-A3B (MoE)
41.9 GB
Qwen 3.5 122B-A10B (MoE)
145.5 GB
Q6_K
Qwen 3.5 35B-A3B (MoE)
32.4 GB
Qwen 3.5 122B-A10B (MoE)
112.4 GB
Q5_K_M
Qwen 3.5 35B-A3B (MoE)
28.1 GB
Qwen 3.5 122B-A10B (MoE)
97.5 GB
Q4_K_M
Qwen 3.5 35B-A3B (MoE)
24.1 GB
Qwen 3.5 122B-A10B (MoE)
83.4 GB
Q3_K_M
Qwen 3.5 35B-A3B (MoE)
19.0 GB
Qwen 3.5 122B-A10B (MoE)
66.0 GB
Q2_K
Qwen 3.5 35B-A3B (MoE)
15.1 GB
Qwen 3.5 122B-A10B (MoE)
52.3 GB
NVFP4
Qwen 3.5 35B-A3B (MoE)
19.8 GB
Qwen 3.5 122B-A10B (MoE)
68.5 GB
QuantQwen 3.5 35B-A3B (MoE)Qwen 3.5 122B-A10B (MoE)Diff
FP32157.0 GB546.8 GB-71%
BF1678.6 GB273.5 GB-71%
FP1678.6 GB273.5 GB-71%
Q8_041.9 GB145.5 GB-71%
Q6_K32.4 GB112.4 GB-71%
Q5_K_M28.1 GB97.5 GB-71%
Q4_K_M24.1 GB83.4 GB-71%
Q3_K_M19.0 GB66.0 GB-71%
Q2_K15.1 GB52.3 GB-71%
NVFP419.8 GB68.5 GB-71%

Diff is Qwen 3.5 35B-A3B (MoE) relative to Qwen 3.5 122B-A10B (MoE). Green = lower VRAM (fits more GPUs).

Model specifications

SpecQwen 3.5 35B-A3B (MoE)Qwen 3.5 122B-A10B (MoE)
OrgAlibabaAlibaba
Parameters35B122B
ArchitectureMoE (3B active)MoE (10B active)
Context256k tokens256k tokens
Modalitiestext, vision, videotext, vision, video
LicenseApache 2.0Apache 2.0
CommercialYesYes
Released2026-02-242026-02-15
GPUs (native)84 / 11933 / 119

Benchmark scores

BenchmarkQwen 3.5 35B-A3B (MoE)Qwen 3.5 122B-A10B (MoE)
MMLU-Pro85.386.7
GPQA Diamond84.286.6
SWE-bench Verified69.272.0

Green = higher score (better). N/A = not yet available. ~ = inherited from a base model, not independently reported for that release itself.

GPUs that run only Qwen 3.5 35B-A3B (MoE)(51)

GPUs that run only Qwen 3.5 122B-A10B (MoE)(0)

Every GPU that runs Qwen 3.5 122B-A10B (MoE) also runs Qwen 3.5 35B-A3B (MoE).

GPUs that run both natively(33)

Which should you use?

Choose Qwen 3.5 35B-A3B (MoE) if:
  • • You have limited VRAM: it's a smaller model needing 24.1 GB vs 83.4 GB
  • • It's the newer release (2026-02-24 vs 2026-02-15); check the benchmark table above for what actually improved
Choose Qwen 3.5 122B-A10B (MoE) if:
  • • You want maximum capability and have a 84 GB+ GPU
  • • Benchmark quality matters: scores 86.7 vs 85.3 on MMLU-Pro

Frequently asked questions

Which is better, Qwen 3.5 35B-A3B (MoE) or Qwen 3.5 122B-A10B (MoE)?
Qwen 3.5 35B-A3B (MoE) has 35B parameters vs 122B for Qwen 3.5 122B-A10B (MoE), so Qwen 3.5 122B-A10B (MoE) is the larger model. Qwen 3.5 35B-A3B (MoE) is more hardware-efficient, needing 24.1 GB at its Q4_K_M build vs 83.4 GB for Qwen 3.5 122B-A10B (MoE)'s Q4_K_M. Qwen 3.5 35B-A3B (MoE) runs on more GPUs natively (84 vs 33). On MMLU-Pro, Qwen 3.5 122B-A10B (MoE) scores higher (86.7 vs 85.3).
How much VRAM does Qwen 3.5 35B-A3B (MoE) need vs Qwen 3.5 122B-A10B (MoE)?
At 8k context, Qwen 3.5 35B-A3B (MoE) needs approximately 24.1 GB of VRAM at its Q4_K_M build, while Qwen 3.5 122B-A10B (MoE) needs 83.4 GB at its Q4_K_M build. At the largest build each ships, Qwen 3.5 35B-A3B (MoE) requires 78.6 GB (FP16) vs 273.5 GB (FP16) for Qwen 3.5 122B-A10B (MoE).
Can you run Qwen 3.5 35B-A3B (MoE) on the same GPUs as Qwen 3.5 122B-A10B (MoE)?
Yes, 33 GPUs can run both natively in VRAM, including NVIDIA B300 288GB, NVIDIA B200 180GB, NVIDIA H200 141GB. However, 51 GPUs can run Qwen 3.5 35B-A3B (MoE) but not Qwen 3.5 122B-A10B (MoE), and no GPU can run Qwen 3.5 122B-A10B (MoE) without also fitting Qwen 3.5 35B-A3B (MoE).
What is the difference between Qwen 3.5 35B-A3B (MoE) and Qwen 3.5 122B-A10B (MoE)?
Qwen 3.5 35B-A3B (MoE) has 35B parameters (3B active, MoE) with a 256k context window. Qwen 3.5 122B-A10B (MoE) has 122B parameters (10B active, MoE) with a 256k context window.
Which model fits in 24 GB of VRAM, Qwen 3.5 35B-A3B (MoE) or Qwen 3.5 122B-A10B (MoE)?
Neither fits in 24 GB: Qwen 3.5 35B-A3B (MoE) needs 24.1 GB at Q4_K_M and Qwen 3.5 122B-A10B (MoE) needs 83.4 GB at Q4_K_M. Both require a multi-GPU server with 84 GB+ of combined VRAM.
Full Qwen 3.5 35B-A3B (MoE) page →Full Qwen 3.5 122B-A10B (MoE) page →Check your hardware →