Qwen3.6-35B-A3B
Released2026-04-16Size35B · 3B activeArchitecturehybrid linear attnLicenseapache-2.0AA rank#36 of 75
Files and memory at 8K context
| Quant | File | Files | + KV + runtime | Smallest fit | Source |
|---|---|---|---|---|---|
| Q4_K_M ggml-org/Qwen3.6-35B-A3B-GGUF | 19.02 GiB | 1 | 20.2 GiB+ | 24 GB GPU | HF files |
| Q5_K_M AtomicChat/Qwen3.6-35B-A3B-GGUF | 23.03 GiB | 1 | 24.2 GiB+ | 32 GB GPU | HF files |
| Q8_0 unsloth/Qwen3.6-35B-A3B-GGUF | 34.37 GiB | 1 | 35.5 GiB+ | 48 GB GPU | HF files |
| MLX 4-bit mixed precision mlx-community/Qwen3.6-35B-A3B-4bit | 19.00 GiB | 4 | 20.2 GiB+ | 32 GB Mac | HF files |
| MLX 8-bit mixed precision mlx-community/Qwen3.6-35B-A3B-8bit | 35.13 GiB | 8 | 36.3 GiB+ | 64 GB Mac | HF files |
No verified file: FP16. Sizes are exact file bytes (text weights; projectors listed separately). “+” = recurrent state or other terms unknown. Smallest fit adds 0.5 GiB spare; Macs use 75% of memory.
Quality
Artificial Analysis Intelligence Index
18.2
Variant: Qwen3.6 35B A3B (Reasoning) · observed 2026-10-17 · source
Rank #36 of 75 researched 2026 models.
Arena text rating
Unknown
Scores describe the evaluated hosted variant, not each quantized file. AA and Arena are never averaged.
Reported speeds 17
| Hardware | Quant | Backend | Prompt t/s | Gen t/s | Date | Source |
|---|---|---|---|---|---|---|
| M5 Pro (20c GPU) 64 GB | MLX 4bit | MLX | 3,338 | 96.7 | 2026-10-15 | omlx.ai |
| M5 Pro (16c GPU) 48 GB | MLX 4bit | MLX | 2,812 | 89.8 | 2026-10-14 | omlx.ai |
| M4 Pro (20c GPU) 64 GB | MLX 4bit | MLX | 1,025 | 89.7 | 2026-10-11 | omlx.ai |
| M4 Pro (16c GPU) 64 GB | MLX 4bit | MLX | 885 | 84.1 | 2026-10-10 | omlx.ai |
| M4 Max (40c GPU) 64 GB | MLX 4bit | MLX | 2,095 | 129 | 2026-10-09 | omlx.ai |
| M2 Max (30c GPU) 64 GB | MLX 4bit | MLX | 782 | 88 | 2026-10-09 | omlx.ai |
| M1 Max (24c GPU) 32 GB | MLX 4bit | MLX | 574 | 66.8 | 2026-10-08 | omlx.ai |
| M5 Max (40c GPU) 128 GB | MLX 4bit | MLX | 3,113 | 137 | 2026-10-06 | omlx.ai |
| M4 Max (32c GPU) 36 GB | MLX 4bit | MLX | 300 | 9.5 | 2026-10-05 | omlx.ai |
| M2 Max (38c GPU) 96 GB | MLX 4bit | MLX | 939 | 75.9 | 2026-09-30 | omlx.ai |
| M3 Max (40c GPU) 64 GB | MLX 4bit | MLX | 1,056 | 72.2 | 2026-09-29 | omlx.ai |
| Apple M2 Max (30-core GPU, 32 GB) | Q4_K_M | llama.cpp Metal | 671 | 58.6 | 2026-09-22 | llama.cpp #4167 |
| M2 Ultra (60c GPU) 64 GB | MLX 4bit | MLX | 1,197 | 87.6 | 2026-08-18 | omlx.ai |
| M2 Ultra (76c GPU) 128 GB | MLX 4bit | MLX | 1,388 | 89 | 2026-08-13 | omlx.ai |
| Surface Laptop Ultra,128GB,RTX Spark | UD-Q4_K_M + MTP | llama.cpp | — | 96.6 | unknown | source |
| Apple M2 Max (32 GB; GPU cores not stated in the dataset card) | MLX 4bit | MLX | 296 | 86.9 | 2026-09-28 | huggingface.co |
| Apple M2 Max (32 GB; GPU cores not stated in the dataset card) | Q4_K_M | llama.cpp Metal | 685 | 57.8 | 2026-09-28 | huggingface.co |
Reported by the linked sources, not reproduced by LocalFit. All rows in the speed database
Community threads
- Possible issue: completion-only output may contain </think> without a preceding <think> due to chat template prefill Hugging Face · 2026-05-22
- Qwen3.6 support for TRT-LLM? Hugging Face · 2026-05-19
Qualitative signals only; titles are not verified claims.