ModelTape
RTX 3060 12 GB, RTX 4070, RTX 5070, RX 7700 XT

Best LLMs for 12 GB of VRAM

124 open-weights models run on a GPU with 12 GB of VRAM and 32 GB of system RAM, 80 of them fully on the GPU. The strongest by benchmark score: Qwen 3.8 27B, Qwen3.6 27B, Gemma 4 31B IT.

Each row shows the best quantization that fits, where it runs and a speed estimate at a 16K context, assuming a card with about 360 GB/s of memory bandwidth. Bigger mixture-of-experts models can keep their experts in system RAM, which is why some run well above 12 GB. Other sizes: 6 GB · 8 GB · 16 GB · 24 GB · 32 GB.

Ranked by benchmarks

#ModelLabReleasedSizeECIBest quantRunstok/s
1Qwen 3.8 27BQwen2026-08-1428B152.6Q3_K_Msplit GPU / CPU~5
2Qwen3.6 27BQwen2026-04-2128B146.7Q3_K_Msplit GPU / CPU~5
3Gemma 4 31B ITGoogle2026-04-0231B144.3IQ3_XXSsplit GPU / CPU~5
4Qwen 3.6 35B-A3BQwen2026-04-1436B MoE144.0Q6_KGPU + experts in RAM~34
5Qwen3.5-35B-A3BQwen2026-02-2436B MoE144.0Q6_KGPU + experts in RAM~34
6Gemma 4 26B A4BGoogle2026-04-0226B MoE141.6Q6_KGPU + experts in RAM~19
7gpt-oss-120bOpenAI2025-08-05117B MoE140.2IQ2_XXSGPU + experts in RAM~36
8Qwen3.5-9BQwen2026-02-249.7B139.3Q6_Kfully on GPU~27
9Qwen3-30B-A3B-Thinking (Jul 2025)Qwen2025-07-3030B MoE138.3Q6_KGPU + experts in RAM~24
10Qwen3-14BQwen2025-04-2915B136.1IQ4_XSfully on GPU~27
11Qwen3-30B-A3B-Instruct (Jul 2025)Qwen2025-07-2930B MoE135.2Q6_KGPU + experts in RAM~24
12gpt-oss-20bOpenAI2025-08-0521B MoE135.0Q6_KGPU + experts in RAM~23
13Qwen3-30B-A3BQwen2025-04-2931B MoE134.1Q6_KGPU + experts in RAM~27
14Qwen3-8BQwen2025-04-288.2B132.1Q6_Kfully on GPU~31
15Phi-4Microsoft2024-12-1215B129.0Q3_K_Mfully on GPU~29
16Gemma 3 27BGoogle2025-03-1227B127.5IQ3_XXSsplit GPU / CPU~7
17phi-3-small 7.4BMicrosoft2024-04-237.4B116.5Q6_Kfully on GPU~34
18phi-3-medium 14BMicrosoft2024-04-2314B114.3IQ4_XSfully on GPU~28
19Mistral NeMoMistral2024-07-1812B113.9Q4_K_Mfully on GPU~28
20Gemma 2 9BGoogle2024-06-249.2B112.7Q6_Kfully on GPU~28
21Llama 3.1-8BMeta2024-07-238B111.1Q6_Kfully on GPU~32
22phi-3-mini 3.8BMicrosoft2024-04-233.8B110.6Q6_Kfully on GPU~65
23Llama 3-8BMeta2024-04-188B109.4Q6_Kfully on GPU~32
24Gemma 7BGoogle2024-02-217B103.3Q6_Kfully on GPU~36
25Mistral 7B v0.1Mistral2023-09-277B102.8Q6_Kfully on GPU~36
26Qwen-14BQwen2023-09-2414B101.9IQ4_XSfully on GPU~28
27Falcon 2 11BTII2024-05-0911B99.3Q5_K_Mfully on GPU~27
28Llama 2-13BMeta2023-07-1813B96.7Q4_K_Mfully on GPU~26
29StarCoder 2 15BHugging Face2024-02-2015B96.3IQ4_XSfully on GPU~26
30Yi 6B01.AI2023-11-226B94.4Q6_Kfully on GPU~42
31Baichuan2-13BBaichuan2023-09-0613B92.0Q4_K_Mfully on GPU~26
32Qwen-7BQwen2023-09-287B89.2Q6_Kfully on GPU~36
33MPT-30BMosaicML2023-06-2230B85.5IQ3_XXSsplit GPU / CPU~6
34LLaMA-13BMeta2023-02-2413B85.4Q4_K_Mfully on GPU~26
35Baichuan 2-7BBaichuan2023-09-207B83.8Q6_Kfully on GPU~36
36Llama 2-7BMeta2023-07-187B83.4Q6_Kfully on GPU~36
37LLaMA-7BMeta2023-02-247B77.5Q6_Kfully on GPU~36
38Gemma 2BGoogle2024-02-212B76.0Q6_Kfully on GPU~120
39MPT-7BMosaicML2023-05-057B73.7Q6_Kfully on GPU~36
40Falcon-7BTII2023-04-247B73.4Q6_Kfully on GPU~36
41StarCoder 2 7BHugging Face2024-02-207B72.7Q6_Kfully on GPU~36
42XGen-7BSalesforce2023-06-277B70.6Q6_Kfully on GPU~36
43DeepSeek Coder 6.7BDeepSeek2023-11-026.7B67.8Q6_Kfully on GPU~38
44StarCoder 2 3BHugging Face2024-02-223B66.9Q6_Kfully on GPU~81
45Dolly 2.0-12bDatabricks2023-04-1112B55.8Q4_K_Mfully on GPU~29
46Cerebras-GPT-13BCerebras2023-03-2013B49.3Q4_K_Mfully on GPU~26

Newer models without benchmark results yet

ModelLabReleasedSizeECIBest quantRunstok/s
KLM 1.6MinimaLabs2026-11-090.1B–Q6_Kfully on GPU~720
Cola2ByteDance2026-11-068.8B–Q6_Kfully on GPU~29
S5 nanoMaritaca-ai2026-11-024.7B–Q6_Kfully on GPU~54
Raon 1 nanoKRAFTON2026-10-293.5B–Q6_Kfully on GPU~69
SenseNova Loop 9B PreviewSenseNova2026-10-289.7B–Q6_Kfully on GPU~27
LBS 27BinclusionAI2026-10-2627B–Q3_K_Msplit GPU / CPU~6
WeVisOne VL 4BTencent2026-10-224.8B–Q6_Kfully on GPU~52
WeVisOne VL 2BTencent2026-10-222.4B–Q6_Kfully on GPU~97
flintRecursiveSI2026-10-2140B MoE–Q6_KGPU + experts in RAM~54
Ling 3.0 flash SanteinclusionAI2026-10-19127B MoE–IQ2_XXSGPU + experts in RAM~37
Axion 1.0 35BAxion1-02026-10-1635B MoE–Q6_KGPU + experts in RAM~27
NVIDIA Nemotron Labs Dory1.0 4B BF16NVIDIA2026-10-164B–Q6_Kfully on GPU~62
NVIDIA Nemotron Labs Dory1.0 4B MXFP8NVIDIA2026-10-164B–Q6_Kfully on GPU~62
Hy Memory 30B A3BTencent2026-10-1530B MoE–Q6_KGPU + experts in RAM~19
granite switch 4.1 8bIBM2026-10-148.8B–Q6_Kfully on GPU~29
Youtu Parsing OmniTencent2026-10-095.3B–Q6_Kfully on GPU~47
MiniCPM5 2BOpenbmb2026-09-062.5B–Q6_Kfully on GPU~95
tiny aya en thinkerCohere2026-09-023.4B–Q6_Kfully on GPU~73
tiny aya l2 thinkerCohere2026-09-023.4B–Q6_Kfully on GPU~73
Qwen3.8 Flash NextQwen2026-08-24125B MoE–Stratavia Strata~51
Muse GlimmerMeta2026-08-1030B–Q3_K_Msplit GPU / CPU~5
Ling 3.0 tinyinclusionAI2026-08-107.9B MoE–Q6_Kfully on GPU~160
North Micro Vision InstructCohere2026-08-102.5B–Q6_Kfully on GPU~96
Muse Glimmer 30B assistantMeta2026-08-0930B–Q3_K_Msplit GPU / CPU~5
Mage VLMicrosoft2026-07-254.7B–Q6_Kfully on GPU~53
Gemma 4 E4BGoogle2026-07-028B–Q6_Kfully on GPU~32
Gemma 4 E2BGoogle2026-07-025.1B–Q6_Kfully on GPU~49
DiffusionGemma 26B-A4BDeepMind2026-06-1026B MoE–Q6_KGPU + experts in RAM~21
Gemma 4 12BGoogle2026-06-0312B–Q4_K_Mfully on GPU~29
Granite 4.1 30BIBM2026-04-2929B–IQ3_XXSsplit GPU / CPU~5
Granite 4.1 8BIBM2026-04-298.8B–Q6_Kfully on GPU~29
Qwen3.5-4BQwen2026-02-244.7B–Q6_Kfully on GPU~54
Qwen3.5-2BQwen2026-02-242.3B–Q6_Kfully on GPU~100
Qwen3.5-0.8BQwen2026-02-240.9B–Q6_Kfully on GPU~230
Qwen3.5 27BQwen2026-02-2328B–Q3_K_Msplit GPU / CPU~5
Qwen3.5 122B-A10BQwen2026-02-23125B MoE–IQ2_XXSGPU + experts in RAM~32
Qwen3-Coder-NextQwen2026-02-0280B MoE–IQ3_XXSGPU + experts in RAM~47
GLM-4.7-FlashZ.ai2026-01-1931B MoE–Q6_KGPU + experts in RAM~20
Solar Open 100BUpstage2025-12-31103B MoE–IQ2_XXSGPU + experts in RAM~19
VAETKINC AI2025-12-30112B MoE–IQ2_XXSGPU + experts in RAM~21