Hardware FAQ
M4 Max — frequently asked questions
Direct answers to the questions local-LLM enthusiasts ask about M4 Max, drawn from 5 signed runs on llm-speed. Every numerical claim links to a verifiable run permalink at /r/<id>.
What's the fastest LLM on M4 Max?
The fastest measured LLM on M4 Max on llm-speed is qwen3-coder-bench-32k at 113.3 decode tok/s on ollama 0.30.11 (workload chat-short, run r_roktphpc--8). Cite as https://llm-speed.com/r/r_roktphpc--8.
This is the headline decode tokens-per-second across every (model, backend) pairing submitted on M4 Max; faster results may exist on hardware not yet benchmarked, but among signed runs this is the published top.
Can I run a 7B-class model on M4 Max?
No 7B-class model run on M4 Max has been submitted to llm-speed yet, so the canonical "yes/no with a measured tok/s" answer is not currently published.
M4 Max has ~128 GB of VRAM/unified memory, which should fit a 7B model at 4-bit quantization; submit a run with "llm-speed bench --models <hf-id>" to populate this answer.
Can I run a 30B-class model on M4 Max?
No 30B-class model run on M4 Max has been submitted to llm-speed yet, so the canonical "yes/no with a measured tok/s" answer is not currently published.
M4 Max has ~128 GB of VRAM/unified memory, which should fit a 30B model at 4-bit quantization; submit a run with "llm-speed bench --models <hf-id>" to populate this answer.
Can I run a 70B-class model on M4 Max?
No 70B-class model run on M4 Max has been submitted to llm-speed yet, so the canonical "yes/no with a measured tok/s" answer is not currently published.
M4 Max has ~128 GB of VRAM/unified memory, which should fit a 70B model at 4-bit quantization; submit a run with "llm-speed bench --models <hf-id>" to populate this answer.
How does M4 Max compare to M3 Ultra for local LLM inference?
For a head-to-head between M4 Max and M3 Ultra, see the side-by-side comparison page at https://llm-speed.com/vs/m4-max-vs-m3-ultra, which lays out every (model, backend) pair where both rigs have a signed run.
As a single-rig anchor, M4 Max tops out at 113.3 decode tok/s on qwen3-coder-bench-32k via ollama (run r_roktphpc--8); the M3 Ultra top number is on its own /hw/<slug> page so the comparison stays grounded in measured numbers, not extrapolation.
M4 Max vs M3 Ultra (side-by-side) · M3 Ultra leaderboard · M4 Max leaderboard
Which backend is fastest on M4 Max?
The only backend with a signed local run on M4 Max so far is ollama, with a top result of 113.3 decode tok/s on qwen3-coder-bench-32k (run r_roktphpc--8); a multi-backend comparison on M4 Max is not yet published.
Submit a competing run with "llm-speed bench --backends <other-backend>" on a M4 Max machine to populate the comparison.
How much unified memory do I need for local LLM inference on M4 Max?
M4 Max ships with up to 128 GB of unified memory, which sets the ceiling on what fits: a 7B-class model at 4-bit needs roughly 6 GB, a 30B-class model at 4-bit needs roughly 18 GB, and a 70B-class model at 4-bit needs roughly 38 GB.
Long-context workloads add KV-cache pressure on top of the weights — a 32k-token context on a 70B-class model adds another 8-16 GB depending on attention layout, so the practical fit is tighter than the weight-only number.
For exact "did it actually run" evidence on M4 Max, the leaderboard at /hw/m4-max lists every model that has produced a signed result on this hardware.
Is M4 Max worth it for local LLM inference in 2026?
On signed data, M4 Max delivers up to 113.3 decode tok/s on qwen3-coder-bench-32k via ollama (run r_roktphpc--8), which puts it comfortably above interactive-coding thresholds (~80 tok/s) for the published top configuration.
"Worth it" depends on your model class: M4 Max is most useful for 30B-70B class local models where its memory ceiling matters.
For "what fits and how fast", the per-model rows on /hw/m4-max are the honest answer; for cross-rig comparisons, see /vs/m4-max-vs-m3-ultra.
What quantization should I use on M4 Max?
On M4 Max, the only quant with a signed local run is Q4_K_M at 113.3 tok/s on qwen3-coder-bench-32k (run r_roktphpc--8); a multi-quant comparison on M4 Max is not yet published.
Quant choice is a quality-vs-speed tradeoff that this hardware FAQ does not arbitrate; llm-speed publishes hardware-side speed, not output quality. For quality scores, see the model card on Hugging Face and the LMSYS Chatbot Arena.