Skip to content
localllm
日本語

Apple Silicon (Mac)

Apple M2

Data as of

Memory bandwidth
100 GB/s
Source
Checked:

With 8 GB

One entry per model, up to 30, sorted by popularity. For each model we show the largest (most precise) quantization that fits in this memory with a context of at least 8K tokens. The context length is the longest one that quantization fits (for models whose maximum is under 8K, such as embedding models, that maximum). Models that are not for chat or text generation are labelled with their use. "Left" is the memory minus the estimate; under 5% of the memory is marked "tight" (not a guarantee that it fits).

  • Qwen3.8-27B UD-IQ1_M
    Context length 8,192 Estimated 8.47 GB 0.12 GB left (tight) Speed 8.9–12
  • Ornith-1.5-9B Q5_K_M
    Context length 16,384 Estimated 8.49 GB 0.10 GB left (tight) Speed 8.8–12
  • gemma-4-12B-it IQ4_XS Image input
    Context length 8,192 Estimated 8.49 GB 0.10 GB left (tight) Speed 8.8–12
  • Ornith-1.0-9B Q5_K_M
    Context length 16,384 Estimated 8.13 GB 0.46 GB left Speed 9.3–12
  • mxbai-embed-large-v1 F32 Embedding (not for chat)
    Context length 512 Estimated 2.41 GB 6.18 GB left Speed 45–60
  • Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP Q4_K_M Image input
    Context length 8,192 Estimated 8.35 GB 0.24 GB left (tight) Speed 8.6–11
  • MOSS-Music-8B-Instruct Q4_K_M
    Context length 16,384 Estimated 8.45 GB 0.14 GB left (tight) Speed 12–15
  • Qwen3-VL-8B-Instruct UD-Q5_K_XL Image input
    Context length 8,192 Estimated 7.94 GB 0.65 GB left Speed 10–14
  • Qwen3.5-9B Q5_K_M Image input
    Context length 8,192 Estimated 8.48 GB 0.11 GB left (tight) Speed 8.4–11
  • Ornith-1.5-9B-uncensored Q5_K_M
    Context length 16,384 Estimated 8.10 GB 0.49 GB left Speed 9.3–12
  • Qwen3.5-4B UD-Q8_K_XL Image input
    Context length 32,768 Estimated 8.12 GB 0.47 GB left Speed 10–13
  • Wan2.1 Q4_K_M
    Context length 16,384 Estimated 8.08 GB 0.51 GB left Speed 10–14
  • Qwen3-ASR-1.7B F16 Speech / audio
    Context length 16,384 Estimated 6.82 GB 1.77 GB left Speed 15–20
  • gemma-4-E4B-it-qat-q4_0 Q4_0 Image input
    Context length 131,072 Estimated 8.42 GB 0.17 GB left (tight) Speed 12–16
  • Ternary-Bonsai-2-9B Q8_0 Image input
    Context length 131,072 Estimated 8.07 GB 0.52 GB left Speed 22–30
  • HyperCLOVAX-SEED-Text-Instruct-1.5B-Q4_K_M Q4_K_M
    Context length 65,536 Estimated 8.34 GB 0.25 GB left (tight) Speed 53–71
  • Hy-MT2-7B Q6_K
    Context length 8,192 Estimated 8.04 GB 0.55 GB left Speed 9.7–13
  • Ternary-Bonsai-2-2B Q2_0 Image input
    Context length 262,144 Estimated 4.83 GB 3.76 GB left Speed 110–146
  • MiMo-V2.6-Distill-Qwen-9B Q5_K_M Image input
    Context length 16,384 Estimated 8.51 GB 0.08 GB left (tight) Speed 8.7–12
  • Cyber-Ornith-1.5-9B-OBLITERATED Q5_K_M
    Context length 16,384 Estimated 8.10 GB 0.49 GB left Speed 9.3–12
  • Sharp-Spark-X2.5-4B Q6_K_XL
    Context length 65,536 Estimated 6.95 GB 1.64 GB left Speed 17–22
  • AdaVaR-3B F16
    Context length 32,768 Estimated 8.23 GB 0.36 GB left (tight) Speed 9.7–13
  • Qwen3-8B Q5_K_L
    Context length 8,192 Estimated 8.29 GB 0.30 GB left (tight) Speed 9.6–13
  • Qwen3.5-9B-Kimi-k3-Distilled Q5_K_M
    Context length 16,384 Estimated 8.28 GB 0.31 GB left (tight) Speed 9.0–12
  • Qwen3-4B-Instruct-2507 UD-Q8_K_XL
    Context length 16,384 Estimated 8.32 GB 0.27 GB left (tight) Speed 12–16
  • B1-9B Q5_K_M
    Context length 16,384 Estimated 8.28 GB 0.31 GB left (tight) Speed 9.0–12
  • Qwen3-VL-4B-Instruct UD-Q8_K_XL Image input
    Context length 16,384 Estimated 8.32 GB 0.27 GB left (tight) Speed 12–16
  • Qwen2.5-7B-Instruct Q6_K_L
    Context length 16,384 Estimated 8.31 GB 0.28 GB left (tight) Speed 9.2–12
  • Qwen3-4B UD-Q8_K_XL
    Context length 16,384 Estimated 8.32 GB 0.27 GB left (tight) Speed 12–16
  • Qwen-Image-2.1-Turbo-Uncensored Q5_K Image / video generation
    Context length 8,192 Estimated 8.34 GB 0.25 GB left (tight) Speed 9.6–13

With 16 GB

One entry per model, up to 30, sorted by popularity. For each model we show the largest (most precise) quantization that fits in this memory with a context of at least 8K tokens. The context length is the longest one that quantization fits (for models whose maximum is under 8K, such as embedding models, that maximum). Models that are not for chat or text generation are labelled with their use. "Left" is the memory minus the estimate; under 5% of the memory is marked "tight" (not a guarantee that it fits).

With 24 GB

One entry per model, up to 30, sorted by popularity. For each model we show the largest (most precise) quantization that fits in this memory with a context of at least 8K tokens. The context length is the longest one that quantization fits (for models whose maximum is under 8K, such as embedding models, that maximum). Models that are not for chat or text generation are labelled with their use. "Left" is the memory minus the estimate; under 5% of the memory is marked "tight" (not a guarantee that it fits).

Frequently asked questions

Which local LLMs run on Apple M2?
On this site, 944 models have a quantization that fits in 24 GB at 8,192 tokens (estimated). Popular ones include Qwen3.8-27B, Ornith-1.5-9B, Huihui-Qwen3.6-35B-A3B-abliterated-MTP, Ornith-1.5-35B-A3B, Huihui-Qwen3.8-27B-abliterated.
What is the largest model that runs on Apple M2?
Among the models listed here, Ling-3.0-flash (about 127.5B); Q1_0 fits in 24 GB (estimated at 8,192 tokens).
How fast does Apple M2 generate text?
With 100 GB/s of memory bandwidth, Qwen3.8-27B at Q6_K (23.9 GB) runs at an estimated 2.5–3.4 tokens per second (60–80% of bandwidth ÷ file size; prompt processing not included).