Can the Apple M3 Ultra (512 GB) run gpt-oss 20B?

Yes, it runs great

Runs great. At F16 it needs about 73.3 GB of 384 GB usable, with an estimated speed of 45.3 tok/s.

Best quant: F16
Generation speed: 45.3 tok/s
Prompt processing: 4,891 tok/s
First token (1K prompt): 0.3 s
Largest safe context: 128K

Best app to run it on the Apple M3 Ultra (512 GB)

On macOS

LM Studio

Desktop app · runs on llama.cpp, MLX · use the MLX build

Runs the MLX build, typically 35–90% faster than GGUF for a model with 3.6B active parameters on Apple silicon.
Point-and-click: search, download and chat in one window.

Open gpt-oss 20B in LM Studio

MLX LM Runs the MLX build, typically 35–90% faster than GGUF for a model with 3.6B active parameters on Apple silicon.
Ollama One command with Ollama’s own build of this model.

Want an agent? gpt-oss 20B supports tool use, so you can pair a llama.cpp or MLX LM server with one of these. Apps that run on this computer come first.

Coding agents

Personal assistants

Reverse engineering? Serve gpt-oss 20B locally and use it from Ghidra, IDA Pro, Binary Ninja, radare2 or JADX with one of these:

How we choose: the files published for this model, your operating system and GPU, whether it fits fully in memory, and published speed tests (MLX vs llama.cpp on Apple silicon, vLLM vs llama.cpp vs Ollama on NVIDIA). Details.

Download gpt-oss 20B

Recommended for the Apple M3 Ultra (512 GB): F16. Every other quantization has its own download link in the table below.

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

ollama run gpt-oss:20b

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

Download the F16 file (opens in a new tab)

huggingface.co/ggml-org/gpt-oss-20b-GGUF (opens in a new tab)

ollama run hf.co/ggml-org/gpt-oss-20b-GGUF:BF16

Original weights · Hugging Face

Full-precision files from OpenAI, for fine-tuning, vLLM, Transformers.

huggingface.co/openai/gpt-oss-20b (opens in a new tab)

No sign-in needed.

Official repo · GitHub

Reference code, inference scripts and docs from OpenAI. The weights themselves download from Hugging Face.

github.com/openai/gpt-oss (opens in a new tab)

git clone https://github.com/openai/gpt-oss

Links checked 2 days ago against Hugging Face, Ollama and GitHub.

How it’s rated

Quality 70Great quality (A) · 70 of 100: how capable the model is at F16, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.: How capable the model itself is at F16, from its editorial quality tier scaled by the quantization or precision. It’s the same on any hardware that runs this variant. Scale: A+ 85+ · A 70+ · B 55+ · C 40+ · D under 40.
Runs Runs great: How well it runs on the Apple M3 Ultra (512 GB): speed, capped by how well it fits in memory. Runs great: 30+ tokens/s (or near the image, video or decision speed target) · Runs well: 10+ tokens/s · Slow: 3+ tokens/s · Very slow: under 3 · Won’t run: doesn’t fit even with offload. A tight fit or RAM offload caps it at Runs well; heavy offload caps it at Slow.

Match score

To order lists, each result also gets a match score out of 100 from four parts. It favors the most capable model that still runs well.

Model qualityEditorial quality tier of the model, scaled by the quantization or precision used.	31.5 / 45
Memory fitFull marks when the model fits in the share of unified memory the GPU can use; fewer when it goes past that limit.	25.0 / 25
SpeedGeneration speed against a 40 tokens/s target for LLMs, or render time against a target for image and video.	15.0 / 15
Memory useRewards using the memory you have, so a bigger, better variant beats a tiny one that leaves most memory idle.	4.4 / 15
Match score	75.9 / 100

The match score orders results; it is not shown as a grade. Quality and Runs are the two ratings in the tables.

Every quantization on the Apple M3 Ultra (512 GB)

Quant	Memory needed	Available	Speed	Runs	Quality	Download
`MXFP4`	43.5 GB	384 GB	138 tok/s	Runs great	70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab) Original weights (opens in a new tab)
`F16`	73.3 GB	384 GB	45.3 tok/s	Runs great	70Great quality (A) · 70 of 100: how capable the model is at F16, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab) Original weights (opens in a new tab)

Memory figures assume 8K of context. Longer conversations or documents grow the KV cache.

Other hardware that runs it

70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. AMD Radeon RX 9070 XT MXFP4 · 120 tok/s · Runs great
70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 4060 Ti 16GB MXFP4 · 77.6 tok/s · Runs great
70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 4070 Ti SUPER MXFP4 · 134 tok/s · Runs great
70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 4080 SUPER MXFP4 · 141 tok/s · Runs great
70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 5060 Ti 16GB MXFP4 · 105 tok/s · Runs great
70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 5070 Ti MXFP4 · 156 tok/s · Runs great

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.