Best LLMs for 24GB VRAM

24GB unlocks 30B-class models at Q4 and 13B models at high quality. With 24GB you can also push context lengths past 32K. Top picks below are ranked against a 24GB reference card.

  1. 1

    Gemma 3 27B

    27B paramsS grade

    Google's flagship open model. Near GPT-4 quality. Needs 20GB+ RAM.

    Min VRAM: 15.91GBQuant: Q4_K_MSize: 15.41GBLicense: gemma
  2. 2

    Mistral Small 22B

    22B paramsS grade

    22B parameter model. Strong reasoning and multilingual. Needs 16GB+ RAM.

    Min VRAM: 12.93GBQuant: Q4_K_MSize: 12.425GBLicense: apache-2.0
  3. 3

    Phi-4

    14B paramsS grade

    Microsoft's 14B parameter model. Punches well above its weight on reasoning.

    Min VRAM: 15.01GBQuant: Q8_0Size: 14.51GBLicense: mit
  4. 4

    Qwen 2.5 14B

    14B paramsS grade

    Strong 14B model with excellent coding and reasoning. iPad Pro recommended.

    Min VRAM: 15.12GBQuant: Q8_0Size: 14.623GBLicense: apache-2.0
  5. 5

    Gemma 3 12B

    12B paramsS grade

    High quality 12B model. Excellent for iPad Pro and Mac.

    Min VRAM: 12.15GBQuant: Q8_0Size: 11.651GBLicense: gemma
  6. 6

    Mistral Nemo 12B

    12B paramsS grade

    Mistral's 12B model with excellent instruction following.

    Min VRAM: 12.63GBQuant: Q8_0Size: 12.128GBLicense: apache-2.0
  7. 7

    Solar 10.7B

    10.7B paramsS grade

    Depth-upscaled 10.7B model. Strong reasoning.

    Min VRAM: 11.12GBQuant: Q8_0Size: 10.621GBLicense: apache-2.0
  8. 8

    Falcon 3 10B

    10B paramsS grade

    10B Falcon model. Good iPad model.

    Min VRAM: 10.7GBQuant: Q8_0Size: 10.203GBLicense: apache-2.0
  9. 9

    Gemma 2 9B Instruct

    9.2B paramsS grade

    Google's efficient 9B model. Great performance-to-size ratio.

    Min VRAM: 9.65GBQuant: Q8_0Size: 9.152GBLicense: gemma
  10. 10

    Yi 1.5 9B Chat

    9B paramsS grade

    9B bilingual model with strong reasoning.

    Min VRAM: 9.24GBQuant: Q8_0Size: 8.739GBLicense: apache-2.0
  11. 11

    DeepSeek R1 Distill 8B

    8B paramsS grade

    Compact reasoning model. Good reasoning capabilities in a small package.

    Min VRAM: 8.45GBQuant: Q8_0Size: 7.954GBLicense: mit
  12. 12

    Llama 3.1 8B Instruct

    8B paramsS grade

    Meta's 8B parameter instruction-tuned model. Great balance of performance and efficiency for local deployment.

    Min VRAM: 8.45GBQuant: Q8_0Size: 7.954GBLicense: llama3.1
  13. 13

    Granite 3.3 8B

    8B paramsS grade

    IBM's 8B instruction model. Enterprise quality.

    Min VRAM: 8.59GBQuant: Q8_0Size: 8.088GBLicense: apache-2.0
  14. 14

    EXAONE 3.5 7.8B

    7.8B paramsS grade

    7.8B model from LG. Strong bilingual Korean/English.

    Min VRAM: 8.24GBQuant: Q8_0Size: 7.741GBLicense: other
  15. 15

    InternLM 2.5 7B

    7.7B paramsS grade

    Strong 7B model from China. Good at tool use and math.

    Min VRAM: 8.16GBQuant: Q8_0Size: 7.659GBLicense: apache-2.0

Related

All models for 24GB VRAMRTX 4090 compatibility