Qwen3.6's 27B dense model (released 2026-04-22), Apache-2.0. Positioned for coding and agentic work at the 27B scale, with a vision encoder and hybrid attention (Gated DeltaNet plus periodic full attention).
0
Quantized builds
40
Benchmark records
1
Discussions
Quantized builds
Community and official quantized releases; "linked records" counts only benchmarks tied to that exact release.
No quantized builds cataloged yet.
Benchmark data
39 configurations
| Hardware | Framework | Quant | Decode | VRAM used | Samples | Evidence |
|---|---|---|---|---|---|---|
| NVIDIA RTX Spark | llama.cpp | FP16 | 5.5 tok/s | 107 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | llama.cpp | Q5_K_M | 8.4 tok/s | 54.83 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | llama.cpp | Q8_0 | 7.7 tok/s | 67.3 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | Ollama | Q5_K_M | 8.8 tok/s | 54.83 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | Ollama | Q8_0 | 6.7 tok/s | 67.3 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 3090 | vLLM | IQ1_S | 96.1 tok/s | 22.8 GB | 1 | L2 Cross-framework verified |
| AMD Ryzen AI Max+ 395 | llama.cpp | IQ3_XS | 27.7 tok/s | 37.25 GB | 1 | L2 Cross-framework verified |
| AMD Ryzen AI Max+ 395 | Ollama | FP16 | 12 tok/s | 107 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | llama.cpp | IQ1_S | 83.4 tok/s | 25.91 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | Ollama | IQ4_XS | 48.6 tok/s | 44.06 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | llama.cpp | IQ3_XS | 91.1 tok/s | 37.25 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | llama.cpp | Q4_K_S | 81.1 tok/s | 46.33 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | Ollama | IQ3_S | 85.9 tok/s | 39.52 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | Ollama | Q6_K | 53.3 tok/s | 58.8 GB | 1 | L2 Cross-framework verified |
| NVIDIA H100 80GB HBM3 | vLLM | Q8_0 | 234.2 tok/s | 67.3 GB | 1 | L2 Cross-framework verified |
| NVIDIA H100 80GB HBM3 | Strata | Q3_K_S | 335.3 tok/s | 38.96 GB | 1 | L2 Cross-framework verified |
| NVIDIA H100 80GB HBM3 | Strata | Q8_0 | 297.9 tok/s | 67.3 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | vLLM | Q3_K_S | 48.5 tok/s | 38.96 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | Ollama | Q4_K_M | 33 tok/s | 48.59 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | llama.cpp | IQ1_S | 267.3 tok/s | 25.91 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | llama.cpp | Q8_0 | 105.8 tok/s | 67.3 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | vLLM | IQ3_S | 220.1 tok/s | 39.52 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | llama.cpp | Q4_K_M | 67.4 tok/s | 48.59 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | Ollama | Q3_K_L | 80.8 tok/s | 42.92 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | llama.cpp | IQ3_S | 13.8 tok/s | 39.52 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | llama.cpp | Q4_K_M | 10.9 tok/s | 48.59 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | Ollama | Q4_K_M | 10.4 tok/s | 48.59 GB | 2 | L2 Cross-framework verified |
| NVIDIA RTX 3090 | Ollama | IQ1_S | 76.6 tok/s | 22.8 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 5090 | Ollama | IQ2_M | 234.2 tok/s | 30.4 GB | 1 | L2 Cross-framework verified |
| Apple M3 Max | Ollama | IQ2_XS | 109.2 tok/s | 29.32 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 5090 | llama.cpp | IQ1_S | 273.1 tok/s | 25.91 GB | 1 | L2 Cross-framework verified |
| Apple M4 Max | llama.cpp | IQ2_M | 189.9 tok/s | 31.58 GB | 1 | L2 Cross-framework verified |
| Apple M4 Max | llama.cpp | IQ2_XS | 191.2 tok/s | 29.32 GB | 1 | L2 Cross-framework verified |
| Apple M4 Max | vLLM | Q3_K_M | 174.2 tok/s | 40.66 GB | 1 | L2 Cross-framework verified |
| Apple M4 Max | Ollama | Q6_K | 99.5 tok/s | 58.8 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 3090 | llama.cpp | Q4_K_M | 57 tok/s | 23.44 GB | 1 | L0 Self-reported |
| NVIDIA RTX 3090 | llama.cpp | UD-Q3_K_XL | 36.7 tok/s | 20.11 GB | 1 | L0 Self-reported |
| NVIDIA RTX 3090 | llama.cpp | Q5_K_S | 46.6 tok/s | 21.53 GB | 1 | L0 Self-reported |
| NVIDIA RTX 3090 Ti | llama.cpp | Q5_K_S | 50.8 tok/s | 22.47 GB | 1 | L0 Self-reported |
40 of these records are not linked to a specific quantized build; the full set is on the search page.