RunLLMLocal
Open the checker

Compare · Models

Gemma 4 26B-A4B (MoE) vs Qwen3.6 35B-A3B (MoE): which runs better locally?

Gemma 4 26B-A4B (MoE) needs about 18.9 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3.6 35B-A3B (MoE) needs 23.5 GB and fits on 1. On cards that run both, Qwen3.6 35B-A3B (MoE) is about 32% faster. Qwen3.6 35B-A3B (MoE) has more parameters (35B vs 25.8B), which usually means better answers if you have the memory.

VerdictGemma 4 26B-A4B (MoE) is easier to run

Gemma 4 26B-A4B (MoE) needs about 18.9 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3.6 35B-A3B (MoE) needs 23.5 GB and fits on 1. On cards that run both, Qwen3.6 35B-A3B (MoE) is about 32% faster. Qwen3.6 35B-A3B (MoE) has more parameters (35B vs 25.8B), which usually means better answers if you have the memory.

Gemma 4 26B-A4B (MoE)Chat, Vision, Reasoning
Parameters25.8B (4B active)
Memory at Q4, 8K18.9 GB
Memory at Q8, 8K32.2 GB
GPUs that run it fully4 / 40
Cheapest desktop cardRTX 3090 / 3090 Ti (24 GB)
Qwen3.6 35B-A3B (MoE)Chat, Coding, Reasoning, Vision
Parameters35B (3B active)
Memory at Q4, 8K23.5 GB
Memory at Q8, 8K41.6 GB
GPUs that run it fully1 / 40
Cheapest desktop cardRTX 5090 (32 GB)

Each GPU at 8K context with the best quantization that fits. Offload assumes 32 GB of system RAM. Green marks the better result. Rows where the two differ come first.

GPUGemma 4 26B-A4B (MoE)Qwen3.6 35B-A3B (MoE)
RTX 4090 24 GBRuns~120 tok/sQ4_K_MOffload~20 tok/sQ4_K_M
RX 7900 XTX 24 GBRuns~110 tok/sQ4_K_MOffload~19 tok/sQ4_K_M
RTX 3090 / 3090 Ti 24 GBRuns~110 tok/sQ4_K_MOffload~19 tok/sQ4_K_M
RTX 5090 32 GBRuns~150 tok/sQ6_KRuns~200 tok/sQ5_K_M
RX 7900 XT 20 GBOffload~14 tok/sQ4_K_MOffload~16 tok/sQ4_K_M
RTX 5080 16 GBOffload~16 tok/sQ4_K_MOffload~17 tok/sQ4_K_M
RTX 5070 Ti 16 GBOffload~15 tok/sQ4_K_MOffload~16 tok/sQ4_K_M
RTX 4080 / 4080 Super 16 GBOffload~13 tok/sQ4_K_MOffload~14 tok/sQ4_K_M
RTX 4070 Ti Super 16 GBOffload~12 tok/sQ4_K_MOffload~14 tok/sQ4_K_M
RX 9070 XT 16 GBOffload~11 tok/sQ4_K_MOffload~13 tok/sQ4_K_M
RX 7800 XT 16 GBOffload~11 tok/sQ4_K_MOffload~13 tok/sQ4_K_M
Arc A770 16 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RX 6800 XT 16 GBOffload~9.7 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 5060 Ti 16 GBOffload~8.8 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RTX 4060 Ti 16 GBOffload~6.7 tok/sQ4_K_MOffload~8.8 tok/sQ4_K_M
RTX 5070 12 GBOffload~12 tok/sQ4_K_MOffload~14 tok/sQ4_K_M
RTX 4070 / 4070 Super 12 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
Arc B580 12 GBOffload~9.7 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RX 7700 XT 12 GBOffload~9.4 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RX 6700 XT 12 GBOffload~8.9 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RTX 3060 12 GBOffload~8.7 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
GTX 1080 Ti 11 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 3080 10 GBOffload~13 tok/sQ4_K_MOffload~14 tok/sQ4_K_M
Arc A750 8 GBOffload~11 tok/sQ4_K_MOffload~13 tok/sQ4_K_M
RTX 5060 Ti 8 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 5060 8 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 3070 / 3070 Ti 8 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 3060 Ti 8 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 4060 Ti 8 GBOffload~9.2 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RX 7600 8 GBOffload~9.2 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RTX 4060 8 GBOffload~9.1 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RTX 3050 8 GBOffload~8.8 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
GTX 1660 / 1660 Super 6 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
GTX 1650 4 GBOffload~10 tok/sQ4_K_MOffload~14 tok/sQ3_K_M
RTX 4090 Laptop 16 GBOffload~11 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 4080 Laptop 12 GBOffload~9.4 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RTX 4070 Laptop 8 GBOffload~9.0 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RTX 4060 Laptop 8 GBOffload~9.0 tok/sQ4_K_MOffload~11 tok/sQ4_K_M
RTX 3060 Laptop 6 GBOffload~10 tok/sQ4_K_MOffload~12 tok/sQ4_K_M
RTX 4050 Laptop 6 GBOffload~9.4 tok/sQ4_K_MOffload~11 tok/sQ4_K_M

Which should you run?

Gemma 4 26B-A4B (MoE) is easier to run. Gemma 4 26B-A4B (MoE) needs about 18.9 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3.6 35B-A3B (MoE) needs 23.5 GB and fits on 1. On cards that run both, Qwen3.6 35B-A3B (MoE) is about 32% faster. Qwen3.6 35B-A3B (MoE) has more parameters (35B vs 25.8B), which usually means better answers if you have the memory.

Only the Gemma 4 26B-A4B (MoE) fits on RTX 4090, RTX 3090 / 3090 Ti and RX 7900 XTX.

See every result on its own page: Gemma 4 26B-A4B (MoE) and Qwen3.6 35B-A3B (MoE).

Questions

Is Gemma 4 26B-A4B (MoE) or Qwen3.6 35B-A3B (MoE) better?

Gemma 4 26B-A4B (MoE) is easier to run. Gemma 4 26B-A4B (MoE) needs about 18.9 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3.6 35B-A3B (MoE) needs 23.5 GB and fits on 1. On cards that run both, Qwen3.6 35B-A3B (MoE) is about 32% faster. Qwen3.6 35B-A3B (MoE) has more parameters (35B vs 25.8B), which usually means better answers if you have the memory. The right pick depends on what you use it for: compare the strengths listed on each model's page.

Which needs less VRAM, Gemma 4 26B-A4B (MoE) or Qwen3.6 35B-A3B (MoE)?

Gemma 4 26B-A4B (MoE) needs about 19 GB at Q4 with 8K context, Qwen3.6 35B-A3B (MoE) about 24 GB. At Q8 it is 32 GB versus 42 GB.

Can I run Gemma 4 26B-A4B (MoE) and Qwen3.6 35B-A3B (MoE) on a 12 GB GPU?

On an RTX 3060 12 GB, Gemma 4 26B-A4B (MoE) only runs with part of the model in system RAM, at a few tokens per second and Qwen3.6 35B-A3B (MoE) only runs with part of the model in system RAM, at a few tokens per second.

More comparisons

Qwen3.6 35B-A3B (MoE) vs Qwen3 30B-A3B (MoE)Model comparisonQwen3 14B vs Gemma 4 12BModel comparisonQwen3 8B vs Llama 3.1 8BModel comparisongpt-oss 20B vs Qwen3 30B-A3B (MoE)Model comparisonQwen3 30B-A3B (MoE) vs Qwen3 32BModel comparisonGemma 4 31B vs Qwen3.6 27BModel comparison

Estimates from memory size and bandwidth at 8K context. Real speed varies with software, drivers and settings.

Copied