Model Spend Arena2ND ED.
27B · dense · 2026-08-14

Qwen3.6 27B

Runs on ≤32 GB · Hugging Face ↗

llama.cppvLLMOllamaLM Studio

Coding index 53.7 (Artificial Analysis)

The strongest measured model that fits a single 24-32 GB card — a dense 27B and the full-precision parent of the ternary Bonsai build.

Sizes on disk

Real GGUF file sizes = weight VRAM.

QuantizationSize
Q2_K12.0 GB
Q3_K_M13.8 GB
IQ4_XS15.7 GB
Q4_016.1 GB
Q4_K_S16.1 GB
Q4_K_M17.1 GB
Q4_117.5 GB
Q417.9 GB
Q5_K_M19.8 GB
Q8_029.0 GB
Q835.8 GB
Q6_K48.9 GB
BF1654.7 GB

Config tips

~18 GB at Q4 on a 24 GB (RTX 4090) or 32 GB card. Native long context; use YaRN to push further. The sensible local flagship for most people.