Yes — with 142.8 GB to spare

Qwen3.5 0.8B at Q4_K_M fits your M2 Ultra · 192 GB entirely in unified memory at 8K context, at an estimated 914 tokens per second. There is room for its full 256K window.

Fully in unified memory 8K context Q4_K_M · 0.5 GB Apache 2.0 Released 28 Feb 2026

Draft model for speculative decoding, or a classifier that fits in 1 GB.

What hardware do I need for Qwen3.5 0.8B? →

The VRAM budget

weights 0.5 GB

Weights 0.5 GB KV cache @ 8K 0.09 GB Runtime overhead 0.6 GB Free 142.8 GB of 144.0 GB

Quantisation ladder

Quant	Weights	Total @ 8K	Max context	Tok/s	Quality	Fit
F16	1.6 GB	2.3 GB	256K	276	Reference	Long context
Q8_0	0.9 GB	1.6 GB	256K	519	−0.1% ppl	Long context
Q6_K	0.7 GB	1.4 GB	256K	673	−0.4% ppl	Long context
Q5_K_M	0.6 GB	1.3 GB	256K	778	−0.8% ppl	Long context
Q4_K_M	0.5 GB	1.2 GB	256K	914	−1.9% ppl	Recommended
Q3_K_M	0.4 GB	1.1 GB	256K	1129	−5.4% ppl	Long context
Q2_K	0.3 GB	1.0 GB	256K	1318	−15% ppl	Long context

Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 6 of its 24 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.

How to run it

Ollama llama.cpp LM Studio MLX

terminal

$ pip install mlx-lm
$ mlx_lm.generate --model mlx-community/Qwen3.5-0.8B-4bit \
    --max-tokens 512 --prompt "Hello"

Apple's own array framework. The fastest path on Apple Silicon. More on MLX.

01Download is 0.5 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.

02macOS caps what the GPU may wire down at about 144.0 GB of 192 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.

03There is room to go to the model's full 256K context on this card.

See all models for this rig Compare with another model