Local AI by memory
Local AI models by memory
Choose how much graphics memory (VRAM) your card has, or how much unified memory your Apple silicon Mac has, to see which documented local configurations in this catalog fit by their publishers' own memory figures. Each page uses the same rule as the finder, names the exact variant, and cites the source of every figure.
Graphics card memory (VRAM)
Compared with publishers' graphics-memory figures only.
- AI models you can run on 8 GB of VRAM3 configurations fit: OpenAI Whisper turbo on your own computer; StarCoder2-3B in bfloat16 with Hugging Face Transformers; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF)
- AI models you can run on 12 GB of VRAM5 configurations fit: OpenAI Whisper turbo on your own computer; StarCoder2-3B in bfloat16 with Hugging Face Transformers; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); StarCoder2-15B in 4-bit with Transformers and bitsandbytes; OpenAI Whisper large-v3 on your own computer
- AI models you can run on 16 GB of VRAM6 configurations fit: OpenAI Whisper turbo on your own computer; StarCoder2-3B in bfloat16 with Hugging Face Transformers; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); StarCoder2-15B in 4-bit with Transformers and bitsandbytes; OpenAI Whisper large-v3 on your own computer; llama.cpp server with Gemma 4 26B A4B (QAT Q4_0 GGUF)
- AI models you can run on 24 GB of VRAM7 configurations fit: OpenAI Whisper turbo on your own computer; StarCoder2-3B in bfloat16 with Hugging Face Transformers; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); StarCoder2-15B in 4-bit with Transformers and bitsandbytes; OpenAI Whisper large-v3 on your own computer; llama.cpp server with Gemma 4 26B A4B (QAT Q4_0 GGUF); llama.cpp server with Muse Glimmer 30B (KQuant-17GB Q4_K_M GGUF)
Apple silicon unified memory (macOS)
Unified memory is compared with graphics-memory and system-memory figures alike.
- AI models you can run on an 8 GB Apple silicon Mac6 configurations fit: llama.cpp with Llama 3.1 8B Instruct quantized to Q4_K_M; LM Studio with Olmo 3 7B Instruct; LM Studio with Rnj-1 Instruct; OpenAI Whisper turbo on your own computer; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); LM Studio with Gemma 4 12B (QAT Q4_0 GGUF)
- AI models you can run on a 16 GB Apple silicon Mac11 configurations fit: llama.cpp with Llama 3.1 8B Instruct quantized to Q4_K_M; LM Studio with Olmo 3 7B Instruct; LM Studio with Rnj-1 Instruct; OpenAI Whisper turbo on your own computer; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); LM Studio with Gemma 4 12B (QAT Q4_0 GGUF); OpenAI Whisper large-v3 on your own computer; llama.cpp server with Gemma 4 26B A4B (QAT Q4_0 GGUF); Cline in VS Code with gpt-oss-20b on Ollama; LM Studio with gpt-oss-20b; Ollama desktop app with gpt-oss-20b
- AI models you can run on a 24 GB Apple silicon Mac13 configurations fit: llama.cpp with Llama 3.1 8B Instruct quantized to Q4_K_M; LM Studio with Olmo 3 7B Instruct; LM Studio with Rnj-1 Instruct; OpenAI Whisper turbo on your own computer; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); LM Studio with Gemma 4 12B (QAT Q4_0 GGUF); OpenAI Whisper large-v3 on your own computer; llama.cpp server with Gemma 4 26B A4B (QAT Q4_0 GGUF); Cline in VS Code with gpt-oss-20b on Ollama; LM Studio with gpt-oss-20b; Ollama desktop app with gpt-oss-20b; LM Studio with Gemma 4 31B (QAT Q4_0 GGUF); llama.cpp server with Muse Glimmer 30B (KQuant-17GB Q4_K_M GGUF)
- AI models you can run on a 32 GB Apple silicon Mac14 configurations fit: llama.cpp with Llama 3.1 8B Instruct quantized to Q4_K_M; LM Studio with Olmo 3 7B Instruct; LM Studio with Rnj-1 Instruct; OpenAI Whisper turbo on your own computer; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); LM Studio with Gemma 4 12B (QAT Q4_0 GGUF); OpenAI Whisper large-v3 on your own computer; llama.cpp server with Gemma 4 26B A4B (QAT Q4_0 GGUF); Cline in VS Code with gpt-oss-20b on Ollama; LM Studio with gpt-oss-20b; Ollama desktop app with gpt-oss-20b; LM Studio with Gemma 4 31B (QAT Q4_0 GGUF); llama.cpp server with Muse Glimmer 30B (KQuant-17GB Q4_K_M GGUF); LM Studio with Muse Glimmer 30B
- AI models you can run on a 64 GB Apple silicon Mac14 configurations fit: llama.cpp with Llama 3.1 8B Instruct quantized to Q4_K_M; LM Studio with Olmo 3 7B Instruct; LM Studio with Rnj-1 Instruct; OpenAI Whisper turbo on your own computer; llama.cpp server with Gemma 4 12B (QAT Q4_0 GGUF); LM Studio with Gemma 4 12B (QAT Q4_0 GGUF); OpenAI Whisper large-v3 on your own computer; llama.cpp server with Gemma 4 26B A4B (QAT Q4_0 GGUF); Cline in VS Code with gpt-oss-20b on Ollama; LM Studio with gpt-oss-20b; Ollama desktop app with gpt-oss-20b; LM Studio with Gemma 4 31B (QAT Q4_0 GGUF); llama.cpp server with Muse Glimmer 30B (KQuant-17GB Q4_K_M GGUF); LM Studio with Muse Glimmer 30B
These are estimates from documentation, not tests, unless an entry says otherwise. They do not cover speed, context length, or every operating system. For a different amount, enter it in the finder or estimate with the model size calculator.
Support Us
Help keep USASI useful.
Optional. No USASI account required. Payment takes place on the linked provider’s website (Buy Me a Coffee).
About supporting this project