Model Spend Arena2ND ED.
9B · dense · 2026-08-14

Qwen3.5 9B

Runs on ≤8 GB · Hugging Face ↗

llama.cppvLLMOllamaLM Studio

Coding index 28.7 (Artificial Analysis)

A dense 9B with a thinking mode — punches above its size on reasoning for something that fits an 8 GB card comfortably.

First-party test · not the AA coding index

HumanEval pass@1 40% (12/30), via local GPU (Ollama, Q4_K_M). Measured by us on this hardware, as a rough sanity check — HumanEval is a different, easier, partly-contaminated benchmark than Artificial Analysis’ composite, so it is not comparable to the coding-index column.

First-party test · not the AA coding index

BigCodeBench-Hard pass@1 12% (14/121), via local GPU (Ollama, num_ctx 16384, budget 12K). The brutal counterpart to HumanEval — where HumanEval saturates near the top, BCB-Hard spreads the field, so this is the number that actually separates coding ability. Comparable full-148 at 16K context, but a very verbose reasoner — 56/148 still hit the 12K budget, so a near-floor.

First-party measurement

~44 tok/s on RTX 4060 Ti, measured by us.

Sizes on disk

Real GGUF file sizes = weight VRAM.

QuantizationSize
Q2_K4.1 GB
Q3_K_M4.7 GB
IQ4_XS5.2 GB
Q4_05.4 GB
Q4_K_S5.4 GB
Q4_K_M5.7 GB
Q4_15.8 GB
Q46.0 GB
Q5_K_M6.6 GB
Q8_09.5 GB
Q813.0 GB
Q6_K16.2 GB
BF1617.9 GB

Config tips

Fits well under 8 GB at Q4. Toggle reasoning with the /think and /no_think controls; disable it for fast, cheap autocomplete-style use.