Qwen3.5-2B
56 records across 11 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| M1 Ultra | Qwen3.5-2B | oQ6e | oMLX | 223.9 | 4343.0 | – | – | 8192 | 2026-10-10 | community | – | source |
| M1 Ultra | Qwen3.5-2B | oQ4e | oMLX | 204.6 | 4700.0 | – | – | 8192 | 2026-10-10 | community | – | source |
| M1 Ultra | Qwen3.5-2B | oQ4e | oMLX | 204.6 | 4679.0 | – | – | 8192 | 2026-10-10 | community | – | source |
| M5 Ultra | Qwen3.5-2B | bf16 | oMLX | 202.6 | 3567.0 | – | – | 1024 | 2026-10-16 | community | – | source |
| M1 Ultra | Qwen3.5-2B | oQ6e | oMLX | 184.3 | 4354.0 | – | – | 8192 | 2026-10-10 | community | contradiction | source |
| M5 Pro | Qwen3.5-2B | oQ3.5 | oMLX | 173.0 | 3745.0 | – | – | 1024 | 2026-10-09 | community | – | source |
| M1 Ultra | Qwen3.5-2B | oQ6e | oMLX | 169.5 | 4357.0 | – | – | 8192 | 2026-10-10 | community | contradiction | source |
| M1 Ultra | Qwen3.5-2B | oQ6e | oMLX | 165.5 | 4331.0 | – | – | 8192 | 2026-10-10 | community | contradiction | source |
| M1 Ultra | Qwen3.5-2B | oQ6e | oMLX | 147.4 | 4371.0 | – | – | 8192 | 2026-10-10 | community | contradiction | source |
| M6 | Qwen3.5-2B | 4bit | oMLX | 126.3 | 4154.0 | – | – | 1024 | 2026-10-14 | community | – | source |
| M4 Pro | Qwen3.5-2B | 6bit | oMLX | 113.3 | 1768.0 | – | – | 1024 | 2026-03-09 | community | – | source |
| M4 Pro | Qwen3.5-2B | 6bit | oMLX | 113.3 | 1768.0 | – | – | 1024 | 2026-03-09 | community | – | source |
| M6 | Qwen3.5-2B | 4bit | oMLX | 113.3 | 4510.0 | – | – | 4096 | 2026-10-15 | community | – | source |
| M4 Pro | Qwen3.5-2B | 6bit | oMLX | 109.1 | 1797.0 | – | – | 4096 | 2026-03-09 | community | – | source |
| M4 Pro | Qwen3.5-2B | 6bit | oMLX | 109.1 | 1797.0 | – | – | 4096 | 2026-03-09 | community | – | source |
| M6 | Qwen3.5-2B | 4bit | oMLX | 101.4 | 4625.0 | – | – | 16384 | 2026-10-15 | community | – | source |
| M5 | Qwen3.5-2B | 4bit | oMLX | 100.8 | 3548.0 | – | – | 1024 | 2026-10-13 | community | – | source |
| M2 Pro | Qwen3.5-2B | OptiQ-4bit | oMLX | 98.8 | 989.9 | – | – | 1024 | 2026-10-06 | community | – | source |
| M5 | Qwen3.5-2B | 4bit | oMLX | 97.8 | 3934.0 | – | – | 4096 | 2026-10-13 | community | – | source |
| M2 Pro | Qwen3.5-2B | OptiQ-4bit | oMLX | 95.0 | 976.9 | – | – | 4096 | 2026-10-06 | community | – | source |
| M4 | Qwen3.5-2B | 4bit | oMLX | 85.5 | 1105.0 | – | – | 1024 | 2026-10-15 | community | – | source |
| M4 | Qwen3.5-2B | 4bit | oMLX | 84.8 | 1105.0 | – | – | 1024 | 2026-10-15 | community | – | source |
| M4 | Qwen3.5-2B | 4bit | oMLX | 83.5 | 931.6 | – | – | 1024 | 2026-10-16 | community | – | source |
| M4 | Qwen3.5-2B | 4bit | oMLX | 81.8 | 1130.0 | – | – | 4096 | 2026-10-15 | community | – | source |
| M4 | Qwen3.5-2B | 4bit | oMLX | 81.7 | 1129.0 | – | – | 4096 | 2026-10-15 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | 81 | – | – | – | tg32 | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | 81 | – | – | – | tg32 | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | 81 | – | – | – | tg32 | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | 81 | – | – | – | tg32 | 2026-10 | community | – | source |
| M4 | Qwen3.5-2B | 4bit | oMLX | 80.5 | 985.1 | – | – | 4096 | 2026-10-16 | community | – | source |
| M5 | Qwen3.5-2B | 4bit | oMLX | 56.7 | 2333.0 | – | – | 65536 | 2026-10-17 | community | – | source |
| M4 | Qwen3.5-2B | OptiQ-4bit | oMLX | 48.2 | 1027.0 | – | – | 1024 | 2026-10-13 | community | – | source |
| M4 | Qwen3.5-2B | OptiQ-4bit | oMLX | 43.4 | 900.0 | – | – | 4096 | 2026-10-13 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | 34.18 | – | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | 33.58 | – | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q2_K | llama.cpp (Hexagon) | 31.86 | – | – | – | – | 2026-10 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q2_K | llama.cpp (Hexagon) | 30.84 | – | – | – | – | 2026-10 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | 30.19 | – | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | 29.53 | – | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q2_K | llama.cpp (Hexagon) | 26.17 | – | – | – | – | 2026-10 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q2_K | llama.cpp (Hexagon) | 25.86 | – | – | – | – | 2026-10 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | 20.68 | – | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | 20.66 | – | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | – | 673.17 | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | – | 673.00 | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | – | 452.70 | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | – | 719.49 | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | – | 445.93 | – | – | 4096 | 2026-10-01 | community | – | source |
| IQ-9075 | Qwen3.5-2B | Q4_0 | llama.cpp (Hexagon) | – | 703.51 | – | – | 4096 | 2026-10-01 | community | – | source |
| Ryzen AI Max+ 395 (Strix Halo) | Qwen3.5-2B | UD-Q4_K_XL | llama.cpp (Vulkan) | – | 213.69 | – | – | 1048576 | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | – | 4084 | – | – | – | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | – | 4092 | – | – | – | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | – | 3117 | – | – | – | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | – | 4161 | – | – | – | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | – | 3506 | – | – | – | 2026-10 | community | – | source |
| Arc Pro B50 | Qwen3.5-2B | Q8_0 | llama.cpp (SYCL) | – | 4335 | – | – | – | 2026-10 | community | – | source |