← AI Hashrate

Best models for 12 GB VRAM

Our catalog lists 7 GPUs with 12 GB VRAM (RTX 5070 12GB, RTX 4070 Ti 12GB, RTX 4070 12GB, RTX 4070 Super 12GB, Arc B580 12GB and 2 more). The ranking uses the RTX 5070 12GB — the highest-bandwidth 12 GB card at 672 GB/s — and shows the 18 of 61 catalog models that fully fit it at Q4 / 8K context (weights + KV cache + 1 GB overhead within 95% of VRAM), fastest first. Methodology.

#ModelParams (B)Q4 tok/sVRAM needed (GB)Fit check
1Qwen3-0.6B0.8534.5 est.1.6fit check
2gemma-3-1b-it1.0427.6 est.1.75fit check
3Llama-3.2-1B-Instruct1.2356.4 est.1.9fit check
4Qwen3-1.7B2.0213.8 est.2.5fit check
5Llama-3.2-3B-Instruct3.2133.6 est.3.4fit check
6Mistral-7B-Instruct-v0.37.2120.5 measured6.4fit check
7Phi-4-mini-instruct3.8112.5 est.3.85fit check
8Phi-4-mini-reasoning3.8112.5 est.3.85fit check
9gemma-3-4b-it4.3109.7 est.4.15fit check
10Qwen3-4B-Instruct-25074.0106.9 est.4.0fit check
11Qwen3-4B-Thinking-25074.0106.9 est.4.0fit check
12Llama-3.1-8B-Instruct8.0100.4 measured7.0fit check
13Qwen2.5-7B-Instruct7.656.3 est.6.7fit check
14Qwen2.5-Coder-7B-Instruct7.656.3 est.6.7fit check
15DeepSeek-R1-Distill-Qwen-7B7.656.3 est.6.7fit check
16Qwen3-8B8.252.2 est.7.15fit check
17DeepSeek-R1-0528-Qwen3-8B8.252.2 est.7.15fit check
18gemma-3-12b-it12.235.6 est.10.11fit check

tok/s are bandwidth-based estimates unless tagged measured. Every 12 GB card fits the same set of models and ranks them identically; absolute tok/s scales with each card's bandwidth.

Buy the RTX 5070 12GB: Search Amazon (affiliate) — Amazon prices change; the price after you click is the live price.

More lists: 7B-class models · 13B-class models · 30B-class models · 70B+ models · Under $500 · Under $1,000 · Under $2,000 · 8 GB VRAM · 16 GB VRAM · 24 GB VRAM · For Qwen3 · For DeepSeek · For Llama · For Gemma · For GLM · For coding · For reasoning · MoE picks