Nemotron 3.5 Super VL 120B-A12B runs at 31 tok/s (30.6 as recorded) on a 128 GB M5 Max, a community run reported by github.com (MLX, NVFP4, 160 generated tokens, greedy). It needs about 72 GB of unified memory. It is NVIDIA's 120B mixture-of-experts model (12B active), released under the OpenMDW licence.
Want to run Nemotron 3.5 Super VL 120B-A12B faster — or step up to bigger models? Find your Mac →This is a community run: a real run, reported by its source and linked below. The index records it as given and never adjusts it.
| Metric | Value |
|---|---|
| Tokens per second | community31 tok/s |
| Provenance | community run · github.com |
| Quantization | NVFP4 |
| Engine | MLX |
| Context | 160 generated tokens, greedy |
| Figure dated | 2026-10-16 |
| Minimum RAM | 72 GB |
| Parameters | 120B MoE (12B active) |
| Chip bandwidth | 614 GB/s |
Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (72 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.
| Speed | Quant | Engine | Mac RAM | Context | Source | Note |
|---|---|---|---|---|---|---|
| community30.6 tok/s | NVFP4 | MLX | 128 GB | 160 generated tokens, greedy | github.com | mlx-vlm PR #2528 (merged 16 Oct) on MLX 0.32.4, MTP off; NVFP4 weights 75.0 GB. 57.1 tok/s with the native MTP head (not a plain-decode figure). |
Ollama — install it (brew install ollama or the app from ollama.com), then pull the model by its tag, checked on ollama.com on 2026-10-17:
MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:
Downloads: Ollama page → · model card →
The 31 tok/s figure above came from MLX at NVFP4; other runtimes and quants run at different speeds.
Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.
| Chip | Speed | Quant | Engine | Bandwidth |
|---|---|---|---|---|
| M5 Ultra | estimated55 tok/s | Q4_K_M | llama.cpp | 1228 GB/s |
| M3 Ultra | estimated39 tok/s | Q4_K_M | llama.cpp | 819 GB/s |
| M4 Max | estimated27 tok/s | Q4_K_M | llama.cpp | 546 GB/s |
NVIDIA's OpenMDW 120B Mamba-Transformer MoE with 12B active (15 Oct 2026): text, image and video in, 1M context; SWE-bench Verified 71.1 on NVIDIA's own NeMo Gym harness. Q4_K_M is ~70 GB (Ollama's nemotron-3.5-super tag is 80–90 GB), so it needs a 96–128 GB Mac; it runs in llama.cpp b11669+, Ollama and mlx-vlm from git, whose PR #2528 reports 30.6 tok/s at NVFP4 on an M5 Max with MTP off. Avoid NVFP4 GGUFs in llama.cpp on a Mac: Metal runs those tensors on the CPU.
OpenMDW licence · 120B MoE · 1000K context · LLMCheck capability 42/50 · see its leaderboard rank →
Nemotron 3.5 Super VL 120B-A12B needs about 72 GB of unified memory. These current Apple Silicon Macs have the headroom to run it comfortably:
As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.
See how Nemotron 3.5 Super VL 120B-A12B stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard