selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

Qwen logo

Can the Apple M3 Ultra (512 GB) run Qwen3.8-Flash-Next?

Qwen3.8-Flash-Next·180B (61.6B active)·Apple M3 Ultra (512 GB)·512 GB unified memory

Saved in this browser, used on every model page.
Yes

Yes. The Apple M3 Ultra (512 GB) runs Qwen3.8-Flash-Next fully in unified memory at FP16 / BF16 (about 347 GB of 384 GB), at roughly 3.7 tokens/s.

best quant
FP16 / BF16
needs
347 GB
usable
384 GB
est. speed
~3.7 tok/s

Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the Qwen3.8-Flash-Next page →

Every quantisation of Qwen3.8-Flash-Next on the Apple M3 Ultra (512 GB)

QuantQualityMemoryFitSpeed
Q2_K Heavy loss59 GBfits~22 tok/s
Q3_K_M Noticeable loss87 GBfits~15 tok/s
Q4_K_M Small loss108 GBfits~12 tok/s
Q5_K_M Near-lossless126 GBfits~10 tok/s
Q6_K Near-lossless146 GBfits~8.9 tok/s
Q8_0 Lossless185 GBfits~6.9 tok/s
FP16 / BF16 pickLossless347 GBfits~3.7 tok/s

Speed is a bandwidth-based estimate for generating text (819 GB/s); prompt processing and long contexts are slower.

Measured speed on the Apple M3 Ultra (512 GB)

No community reports for this pair yet.

Report your own speed →

Run Qwen3.8-Flash-Next on the Apple M3 Ultra (512 GB)

Using FP16 / BF16, the best fit for Apple M3 Ultra (512 GB).

Ollama · From the Ollama librarydocs
ollama run qwen3.8-flash-next

The library build picks its own default quantisation.

Ollama · BF16 straight from Hugging Facedocs
ollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:BF16
llama.cpp · OpenAI-compatible server with a web UIdocs
llama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:BF16

brew install llama.cpp ships with Metal. Downloads the GGUF on first run.

LM Studio · Download with the lms CLIdocs
lms get unsloth/Qwen3.8-Flash-Next-GGUF@BF16

Then load it from the LM Studio app or with `lms load`.

Transformers · Ask about an imagedocs
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="Qwen/Qwen3.8-Flash-Next", dtype="auto", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])

Vision models that fit the Apple M3 Ultra (512 GB) comfortably

All that fit →
ModelParamsBest quantMemorySpeedLicence
220BQ8_0≈3.5 GBFits in memory~1.9 tok/sConditions
321BA18.3BQ8_0≈356 GBFits in memory~23 tok/sCommercial OK
125BA10BQ8_0≈128 GBFits in memory~42 tok/sCommercial OK
236BA22BQ8_0≈246 GBFits in memory~19 tok/sCommercial OK
124BA12BQ8_0≈130 GBFits in memory~35 tok/sCommercial OK
tencent logoHy4-flashtencent
220BQ8_0≈232 GBFits in memory~1.9 tok/sCheck licence

Other GPUs that run Qwen3.8-Flash-Next

Frequently asked questions

Can the Apple M3 Ultra (512 GB) run Qwen3.8-Flash-Next?
Yes. The Apple M3 Ultra (512 GB) runs Qwen3.8-Flash-Next fully in unified memory at FP16 / BF16 (about 347 GB of 384 GB), at roughly 3.7 tokens/s.
How fast does Qwen3.8-Flash-Next run on the Apple M3 Ultra (512 GB)?
An estimate from memory bandwidth gives about 3.7 tokens/s at FP16 / BF16.
Which quantisation of Qwen3.8-Flash-Next should I use on the Apple M3 Ultra (512 GB)?
FP16 / BF16: the largest one that fits. Expected quality: lossless compared with the full-precision weights.

Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.