Best local AI models for 16 GB of VRAM

23 open models load in 16 GB of graphics memory at Q4_K_M with an 8K context; 7 of them have an independent intelligence score. Here they are ranked, with the longest context each can take.

Top picks for 16 GB

#1 · ECI 139.5

Qwen3.5 9B

Memory: 5.8 GB
Max context: 256K
Speed: ~108 t/s

#2 · ECI 137.8

gpt-oss 20b

Memory: 11.3 GB
Max context: 128K
Speed: ~299 t/s

#3 · ECI 133.2

Magistral Small 2506

Memory: 14.9 GB
Max context: 8K
Speed: ~42 t/s

Speeds on RTX 5080, the newest NVIDIA card with this much memory. Max context is the longest that still loads without spilling into system RAM.

Every scored model that fits 16 GB

#	Model	ECI	Memory	Max context	Speed
1	Qwen3.5 9B	139.5	5.8 GB	256K	~108 t/s
2	gpt-oss 20b	137.8	11.3 GB	128K	~299 t/s
3	Magistral Small 2506	133.2	14.9 GB	8K	~42 t/s
4	Mistral Small 3.2 24B	131.7	14.9 GB	8K	~42 t/s
5	Magistral Small 2509	131.4	14.9 GB	8K	~42 t/s
6	Phi-4 14B	130.4	10.1 GB	16K	~65 t/s
7	Llama 3.1 8B	116.6	5.9 GB	64K	~115 t/s