Qwen3.5-9B-MTP
1recorded run
164.3best tok/s
1verified
What the Qwen3.5-9B-MTP numbers mean
The fastest verified result on our board is 164.3 output tok/s on an RTX 3090 at Q6_K_XL with a 24576-token context. That is about 6.1 ms per generated token — 500 tokens takes roughly 3.0 seconds. Prompt processing on the same record runs at 1760.2 tok/s, so a 2,000-token prompt is read in about 1.1 seconds. Peak VRAM was measured at 9.1 GB.
How much VRAM does Qwen3.5-9B-MTP need?
Qwen3.5-9B-MTP has 9.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.
| Quantisation | Estimated VRAM | Smallest card that fits |
|---|---|---|
| Q4_K_M (4-bit) | 6.6 GB | 8 GB |
| Q8_0 (8-bit) | 11.8 GB | 16 GB |
| FP16 (16-bit) | 21.5 GB | 32 GB |
Fastest GPUs for this model (best tok/s)
- RTX 3090164.3
Models with a similar footprint
| Model | Params | Q4_K_M weights |
|---|---|---|
| Carnice-9b | 9B | 5.1 GB |
| DeepSeek-V4-Pro-Qwen3.5-9B-MTP | 9B | 5.1 GB |
| Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated-i1 | 9B | 5.1 GB |
| NVIDIA-Nemotron-Nano-9B-v2 | 9B | 5.1 GB |
| Ornith-1.0-9B | 9B | 5.1 GB |
| Qwen3-VL-8B-Instruct | 9B | 5.1 GB |