Can Mac Studio M3 Ultra 96GB run Qwen3-235B-A22B-Instruct-2507?

Apple · 96 GB LPDDR5 · 819 GB/s bandwidth · Alibaba · 235.1B (active 22B) · MoE · model ctx up to 256K

No — Qwen3-235B-A22B-Instruct-2507 at Q4 needs ≈ 132.5 GB even at 4K context, beyond the Mac Studio M3 Ultra 96GB's 96 GB LPDDR5. It would only run with heavy CPU/disk offload. Decode at Q4/8K: ≈ 34.4 tok/s (estimated, batch 1). FP16 needs ≈ 475.6 GB — does not fit on this card. Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.

Fit & speed by quant and context

Quant	Context	VRAM needed	Fits?	tok/s (decode)
Q4	4K	132.5 GB	No	35.5 est.
Q4	8K default	134.71 GB	No	34.4 est.
Q4	32K	147.91 GB	No	28.5 est.
FP16	4K	473.4 GB	No	—
FP16	8K default	475.6 GB	No	—
FP16	32K	488.8 GB	No	—

Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s means the model is far beyond this card (offload-only territory).

VRAM breakdown at 8K context

Quant	Weights	KV cache	Overhead	Total needed	Mac Studio M3 Ultra 96GB VRAM
Q4	129.31 GB	4.4 GB	1 GB	134.71 GB	96 GB
FP16	470.2 GB	4.4 GB	1 GB	475.6 GB	96 GB

Other GPUs that run Qwen3-235B-A22B-Instruct-2507

• B200 192GB SXM — ≈231.4 tok/s Q4 est.
• MI300X 192GB — ≈153.3 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — ≈23.1 tok/s Q4 est.

All GPUs for Qwen3-235B-A22B-Instruct-2507 →

Other models for the Mac Studio M3 Ultra 96GB

• Qwen3-0.6B — ≈651.5 tok/s Q4 est.
• gemma-3-1b-it — ≈521.2 tok/s Q4 est.
• Llama-3.2-1B-Instruct — ≈434.3 tok/s Q4 est.
• Qwen3-1.7B — ≈260.6 tok/s Q4 est.
• Qwen3-30B-A3B — ≈173.7 tok/s Q4 est.
• Qwen3-Coder-30B-A3B-Instruct — ≈173.7 tok/s Q4 est.
• GLM-4.7-Flash — ≈173.7 tok/s Q4 est.
• Qwen3-30B-A3B-Instruct-2507 — ≈173.7 tok/s Q4 est.

All models on the Mac Studio M3 Ultra 96GB →