VRAM TableGPU fit and local-LLM VRAM for OEM desktops, from published specifications Search by model
Browse categories

Which local language models fit in 4, 8, 12, 16 GB of card memory

Last updated 2026-09-26

Grouped by card memory, because that is the figure a reader knows before buying anything. Each row is the outcome for every card of that memory size in this database; the arithmetic behind an outcome is on each model's own page. Where cards of the same size disagree, the row shows the better outcome and names nothing that a card of that size does not reach.

We have not physically tested this combination. Data is compiled from official specification sheets and cited sources. Verify with the manufacturer before purchasing.

Grouped by card memory. An outcome here is the best any card of that memory size reaches; the arithmetic is on each model's page.
Card memoryCards Fits at 32k context Fits at a short context only Exceeds
96 GB 1 GLM-4.5-Air Estimated · GPT-OSS 120B · Llama 3.1 70B Estimated · Qwen3 32B · EXAONE 4.0 32B · Qwen3 30B-A3B Instruct 2507 Estimated · Gemma 3 27B · Mistral Small 3.2 24B Instruct Estimated · Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B — 0
80 GB 2 GLM-4.5-Air Estimated · GPT-OSS 120B · Llama 3.1 70B Estimated · Qwen3 32B · EXAONE 4.0 32B · Qwen3 30B-A3B Instruct 2507 Estimated · Gemma 3 27B · Mistral Small 3.2 24B Instruct Estimated · Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B — 0
48 GB 6 Qwen3 32B · EXAONE 4.0 32B · Qwen3 30B-A3B Instruct 2507 Estimated · Gemma 3 27B · Mistral Small 3.2 24B Instruct Estimated · Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Llama 3.1 70B Estimated 2
40 GB 1 Qwen3 32B · EXAONE 4.0 32B · Qwen3 30B-A3B Instruct 2507 Estimated · Gemma 3 27B · Mistral Small 3.2 24B Instruct Estimated · Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B — 3
32 GB 4 Qwen3 32B · EXAONE 4.0 32B · Qwen3 30B-A3B Instruct 2507 Estimated · Gemma 3 27B · Mistral Small 3.2 24B Instruct Estimated · Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B — 3
24 GB 17 EXAONE 4.0 32B · Qwen3 30B-A3B Instruct 2507 Estimated · Gemma 3 27B · Mistral Small 3.2 24B Instruct Estimated · Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Qwen3 32B 3
20 GB 4 Gemma 3 27B · Mistral Small 3.2 24B Instruct Estimated · Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Qwen3 30B-A3B Instruct 2507 Estimated 5
16 GB 28 Phi-4 14B Estimated · GPT-OSS 20B · Qwen3 14B · Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Mistral Small 3.2 24B Instruct Estimated 7
12 GB 17 Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Phi-4 14B Estimated · Qwen3 14B 9
11 GB 2 Gemma 3 12B · Qwen3 8B · Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Phi-4 14B Estimated · Qwen3 14B 9
10 GB 3 Llama 3.1 8B Estimated · Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Gemma 3 12B · Qwen3 8B 11
8 GB 43 Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Qwen3 8B · Llama 3.1 8B Estimated 12
6 GB 13 Qwen3 1.7B Estimated · Qwen3 0.6B Estimated · Gemma 3 4B Qwen3 4B 14
5 GB 2 Qwen3 0.6B Estimated · Gemma 3 4B Qwen3 4B · Qwen3 1.7B Estimated 14
4 GB 27 No model fits at this memory size Qwen3 4B · Qwen3 1.7B Estimated · Qwen3 0.6B Estimated 15
3 GB 4 No model fits at this memory size Qwen3 1.7B Estimated · Qwen3 0.6B Estimated 16
2 GB 28 No model fits at this memory size Qwen3 0.6B Estimated 17
1 GB 6 No model fits at this memory size — 18

Power. Leave headroom; OEM PSUs are often rated at the limit.

Every row here is a summary of figures published on the individual pages, where each one carries its source, the date it was read and a confidence grade. Follow the links before acting on a row.

We have not physically tested this combination. Data is compiled from official specification sheets and cited sources. Verify with the manufacturer before purchasing.