Configs › Qwen3.5-4B on RTX 5090: 231.9 tok/s measured

Qwen3.5-4B on RTX 5090: 231.9 tok/s measured

Measured Qwen3.5-4B speeds: 231.9 tok/s on RTX 5090 (UD-IQ3_XXS, llama.cpp/cuda). 19 configs, each linked to its source, snapshot 2026-10-15.

Snapshot 2026-10-15 · 394 configs on AI boxes and 423 on GPU cards · 123 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.

Quick answers

How fast is Qwen3.5-4B on RTX 5090?: 231.9 tok/s single-stream decode (UD-IQ3_XXS, llama.cpp/cuda), the fastest of 19 measured configs on NVIDIA RTX 5090 (32GB), measured by paulplee, source last checked 2026-10-15. source

Qwen3.5-4B on GPU cards, ranked on their own

GPU cards run the model from their own VRAM: very fast on what fits, but most cannot load what a 128 GB AI box can. Ranked on their own, never against the boxes.

Model	Hardware	Quant	Backend	Decode	Context	Trust	Source
Qwen3.5-4B	RTX 5090	UD-IQ3_XXS	llama.cpp/cuda	231.9 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q4_0	llama.cpp/cuda	223.7 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	IQ4_XS	llama.cpp/cuda	220.6 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q4_1	llama.cpp/cuda	217.3 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q4_K_S	llama.cpp/cuda	216.4 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	IQ4_NL	llama.cpp/cuda	215.2 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q4_K_M	llama.cpp/cuda	214.5 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	UD-Q3_K_XL	llama.cpp/cuda	211.6 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	UD-Q4_K_XL	llama.cpp/cuda	208.9 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q5_K_S	llama.cpp/cuda	204.1 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q5_K_M	llama.cpp/cuda	201 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q3_K_S	llama.cpp/cuda	199.6 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q3_K_M	llama.cpp/cuda	199.1 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	UD-Q5_K_XL	llama.cpp/cuda	197.6 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q6_K	llama.cpp/cuda	192.3 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	UD-Q6_K_XL	llama.cpp/cuda	178.9 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	Q8_0	llama.cpp/cuda	176.4 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	UD-Q8_K_XL	llama.cpp/cuda	149 tok/s	8192	Structured table	source
Qwen3.5-4B	RTX 5090	BF16	llama.cpp/cuda	127.1 tok/s	8192	Structured table	source

More: all configs · hardware · news · contributors · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.