community pool / model answer
Qwen3.5-0.8B
chat · Qwen/Qwen3.5-0.8B
445.7 tok/s · model median · 14 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| AMD Ryzen 7 7840HS | 4-bit | 77.8 tok/s | reported | - | - GB | 2,048 | llama.cpp |
| Radeon AI Pro R9700 32GB | 4-bit | 562.2 tok/s | measured | 44 | - GB | 128 | hipfire |
| RTX 3060 12GB | 4-bit | 209.8 tok/s | reported | 59 | - GB | 4,096 | llama.cpp |
| RTX 3060 12GB | 8-bit | 242.8 tok/s | reported | 433 | 3 GB | 4,608 | llama.cpp |
| RTX 5060 Ti 16GB | 4-bit | 280.9 tok/s | reported | 48 | - GB | 4,096 | llama.cpp |
| RTX PRO 6000 Blackwell 96GB | 16-bit | 778.4 tok/s | reported | 37 | - GB | 2,048 | vllm |
| RX 5700 XT 8GB | 4-bit | 273.6 tok/s | reported | 347 | - GB | 128 | hipfire |
| RX 6900 XT 16GB | 4-bit | 346.1 tok/s | reported | 81 | - GB | 128 | hipfire |
| RX 7900 XTX 24GB | 4-bit | 550.2 tok/s | reported | 69 | - GB | 128 | hipfire |
| RX 9070 XT 16GB | 4-bit | 545.3 tok/s | reported | 67 | - GB | 128 | hipfire |
| Ryzen AI Max 395 103GB | 4-bit | 291.4 tok/s | reported | 69 | - GB | 128 | hipfire |
Context tested
The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.