Can I run GPT-oss 20B on a Mac?

Yes: GPT-oss 20B runs on 53 of 54 Macs in the LLMCheck index. It needs about 13 GB at a 4K context, 15 GB at 64K. The smallest and cheapest current Mac at 15+ tok/s is the Mac mini M6 16 GB ($899; ~41 tok/s est.), with the GPU limit raised. Best measured: 184 tok/s on an M5 Ultra (community).

RAM at 4K
13
GB
RAM at 64K
15
GB, 16-bit KV
Macs that run it
53
of 54 configs
Best measured
community184
tok/s · M5 Ultra

GPT-oss 20B on every Mac

One row per Mac configuration, grouped by product line. The build is the one each speed is for; the context column is the longest prompt that fits inside macOS’s default GPU memory limit with a 16-bit KV cache (“raised” where the limit has to be raised first; “full” when the whole window fits). Speeds are decode tok/s from the LLMCheck index, each with its provenance mark. For small mixture-of-experts models (40B total or less), the formula’s typical error is about 15% (the median gap to 88 measured plain-decode runs; it tends to undershoot (median −7%), and an estimate scaled from the model’s own measured run is usually closer; method).

Runs fits macOS’s default GPU limit, 15+ tok/sRuns, slowly fits, under 15 tok/sRaise GPU limit fits once the limit is raised (command below)Doesn't fit needs more memory than the Mac has
MacVerdictBuildSpeedMax contextModel page
MacBook Air
MacBook Air M5 16 GBRaise GPU limit command ↓MXFP4estimated41 tok/s——
MacBook Air M5 24 GBRunsMXFP4estimated41 tok/s128K (full)—
MacBook Air M5 32 GBRunsMXFP4estimated41 tok/s128K (full)—
MacBook Air M1 8 GB refurbDoesn't fit————
MacBook Air M4 16 GB usedRaise GPU limit command ↓MXFP4estimated34 tok/s—M4 page
MacBook Air M2 16 GB refurbRaise GPU limit command ↓MXFP4estimated29 tok/s——
MacBook Air M3 16 GB refurbRaise GPU limit command ↓MXFP4estimated29 tok/s——
MacBook Air M1 16 GB refurbRaise GPU limit command ↓MXFP4estimated21 tok/s——
MacBook Air M4 24 GB usedRunsMXFP4estimated34 tok/s123KM4 page
MacBook Air M2 24 GB refurbRunsMXFP4estimated29 tok/s123K—
MacBook Air M3 24 GB refurbRunsMXFP4estimated29 tok/s123K—
MacBook Pro
MacBook Pro M6 16 GBRaise GPU limit command ↓MXFP4estimated41 tok/s—M6 page
MacBook Pro M5 Pro 24 GBRunsMXFP4estimated69 tok/s128K (full)—
MacBook Pro M6 24 GBRunsunstatedcommunity45 tok/s128K (full)M6 page
MacBook Pro M6 32 GBRunsunstatedcommunity45 tok/s128K (full)M6 page
MacBook Pro M5 Pro 48 GBRunsMXFP4estimated69 tok/s128K (full)—
MacBook Pro M5 Max 64 GBRunsMXFP4community141 tok/s128K (full)M5 Max page
MacBook Pro M5 Pro 64 GBRunsMXFP4estimated69 tok/s128K (full)—
MacBook Pro M5 Max 128 GBRunsMXFP4community141 tok/s128K (full)M5 Max page
MacBook Pro M3 Max 64 GB refurbRunsMXFP4estimated121 tok/s128K (full)—
MacBook Pro M3 Max 128 GB refurbRunsMXFP4estimated121 tok/s128K (full)—
Mac mini
Mac mini M6 16 GBRaise GPU limit command ↓MXFP4estimated41 tok/s—M6 page
Mac mini M5 Pro 24 GBRunsMXFP4estimated69 tok/s128K (full)—
Mac mini M6 24 GBRunsunstatedcommunity45 tok/s128K (full)M6 page
Mac mini M6 32 GBRunsunstatedcommunity45 tok/s128K (full)M6 page
Mac mini M5 Pro 48 GBRunsMXFP4estimated69 tok/s128K (full)—
Mac mini M5 Pro 64 GBRunsMXFP4estimated69 tok/s128K (full)—
Mac mini M2 Pro 16 GB refurbRaise GPU limit command ↓MXFP4estimated51 tok/s——
Mac mini M4 16 GB usedRaise GPU limit command ↓MXFP4estimated34 tok/s—M4 page
Mac mini M4 Pro 24 GB usedRunsMXFP4estimated64 tok/s123KM4 Pro page
Mac mini M4 24 GB usedRunsMXFP4estimated34 tok/s123KM4 page
Mac mini M2 Pro 32 GB refurbRunsMXFP4estimated51 tok/s128K (full)—
Mac mini M4 Pro 48 GB usedRunsMXFP4estimated64 tok/s128K (full)M4 Pro page
Mac mini M4 Pro 64 GB usedRunsMXFP4estimated64 tok/s128K (full)M4 Pro page
iMac
iMac M6 16 GBRaise GPU limit command ↓MXFP4estimated41 tok/s—M6 page
iMac M6 24 GBRunsunstatedcommunity45 tok/s128K (full)M6 page
iMac M6 32 GBRunsunstatedcommunity45 tok/s128K (full)M6 page
Mac Studio
Mac Studio M5 Max 36 GBRunsMXFP4estimated128 tok/s128K (full)M5 Max page
Mac Studio M5 Max 48 GBRunsMXFP4community141 tok/s128K (full)M5 Max page
Mac Studio M5 Max 64 GBRunsMXFP4community141 tok/s128K (full)M5 Max page
Mac Studio M5 Ultra 96 GBRunsMXFP4community184 tok/s128K (full)M5 Ultra page
Mac Studio M5 Max 128 GBRunsMXFP4community141 tok/s128K (full)M5 Max page
Mac Studio M5 Ultra 256 GBRunsMXFP4community184 tok/s128K (full)M5 Ultra page
Mac Studio M5 Ultra 512 GBRunsMXFP4community184 tok/s128K (full)M5 Ultra page
Mac Studio M2 Max 32 GB refurbRunsMXFP4estimated121 tok/s128K (full)—
Mac Studio M1 Ultra 64 GB refurbRunsMXFP4estimated151 tok/s128K (full)—
Mac Studio M4 Max 64 GB usedRunsMXFP4estimated135 tok/s128K (full)M4 Max page
Mac Studio M2 Max 64 GB refurbRunsMXFP4estimated121 tok/s128K (full)—
Mac Studio M1 Ultra 128 GB refurbRunsMXFP4estimated151 tok/s128K (full)—
Mac Studio M2 Ultra 128 GB refurbRunsMXFP4estimated151 tok/s128K (full)—
Mac Studio M4 Max 128 GB usedRunsMXFP4estimated135 tok/s128K (full)M4 Max page
Mac Studio M2 Ultra 192 GB refurbRunsMXFP4estimated151 tok/s128K (full)—
Mac Studio M3 Ultra 256 GB usedRunsMXFP4estimated152 tok/s128K (full)—
Mac Studio M3 Ultra 512 GB refurbRunsMXFP4estimated152 tok/s128K (full)—
Provenance key
sourced
Published figure — vendor or press review
estimated
community
Linked individual run — blog, forum or submission

Context memory is computed from its config.json (12 full layers + 12 sliding-window layers (window 128)): about 1.6 GB of 16-bit KV cache at 64K tokens. A q8 cache (llama.cpp -ctk q8_0 -ctv q8_0, Ollama OLLAMA_KV_CACHE_TYPE=q8_0) halves that part. Runtimes also grow a compute buffer with context, so a fit within about 1 GB of the limit is optimistic.

Raising the GPU memory limit

macOS lets the GPU wire only part of unified memory by default (Metal’s recommended working set; M5 and M6 Macs run macOS 26 or later, whose default is higher), and GPT-oss 20B needs about 13 GB. On the Macs marked Raise GPU limit it fits in RAM but not inside that default. The default is a performance threshold, not a hard wall: past it, the runtime moves part of it to the CPU or shortens the context (llama.cpp, Ollama), runs it slowly (MLX), or stops with an out-of-memory error. One command raises the limit until the next restart (sudo sysctl iogpu.wired_limit_mb=0 restores the default sooner); it leaves macOS a reserve, and other apps get less memory while the model is loaded.

GPT-oss 20B: memory by quantization

Weights are 21B parameters at each format’s effective size (the index’s GB-per-billion table), plus about 2 GB for the runtime and a short context; longer contexts add the KV cache computed from the model’s config.json at 16 bits. It is a mixture-of-experts model: all 21B parameters sit in memory although only 3.6B are read per token. GPT-oss 20B ships natively in MXFP4, which the index costs at 0.54 GB per billion parameters; the other rows are third-party re-quantizations. The first row is the catalog’s own figure (about 13 GB at a 4K context), the one every verdict on this page uses. “Smallest Mac” is the smallest memory size among the Macs Apple sells new that holds it at a 4K context inside that Mac’s default GPU limit. Estimates, not measurements.

BuildWeightsRAM at 4Kat 32Kat 64KSmallest Mac
As catalogued—13 GB14 GB15 GB24 GB
Q2_K7.1 GB9 GB10 GB11 GB16 GB
Q3_K_M10.3 GB12 GB13 GB14 GB24 GB
Q4_K_M12.0 GB14 GB15 GB15 GB24 GB
Q5_K_M14.9 GB17 GB18 GB18 GB24 GB
Q6_K17.2 GB19 GB20 GB21 GB32 GB (24 GB with a raised limit)
Q8_022.3 GB24 GB25 GB26 GB32 GB
F1642.0 GB44 GB45 GB46 GB64 GB

Measured runs of GPT-oss 20B

Every published (filled square) and community (hollow circle) run the index holds for GPT-oss 20B: 3 published, 4 community, each linked to its source. LLMCheck runs no tests of its own. One row is one measurement session; on each chip the table above shows the lower median of the eligible runs (plain decode at 5 bits or fewer), and a binned or “not an anchor” run is shown but never used as a reference.

ChipRuntimeQuantDecodeContextDateSource
M5 Ultra 256 GBMLXMXFP4community184 tok/s512-token random prompt, 128 generated, empty context, mean of 22026-10-15bohemiansilicon.com
M5 Max 128 GBllama.cppMXFP4sourced156 tok/srandom-token prompt, empty context2026-11-02vsesmart.cz
M5 Max 128 GBMLXMXFP4sourced142 tok/srandom-token prompt, empty context2026-11-02vsesmart.cz
M5 Max 128 GBMLXMXFP4community140.5 tok/s1k2026-10-31omlx.ai
M5 Max 128 GBOllamaunstatedsourced107.51 tok/s—2026-09-21tbreak.com
M6 32 GBMLX4-bitcommunity65.6 tok/s1k2026-11-14omlx.ai
M6 32 GBLM Studiounstatedcommunity45 tok/s—2026-10github.com

Recommended setup

The quickest start is Ollama. Pull the model by its tag (checked on ollama.com, 2026-11-14; the file is about 14 GB):

ollama run gpt-oss:20b

Prefer a window to a terminal? LM Studio downloads the same model with one click. · model card →

On the Mac mini M6 (24 GB), the smallest current Mac that runs it inside the default GPU limit, the index figure is 45 tok/s (community run) at unstated; keep the context at or under 128K there, or use a q8 KV cache. The gpt-oss:20b file fits there.

Step-by-step: the Ollama install guide.

🛒 Which Mac to get for GPT-oss 20B

Mac mini M6 (16 GB) — smallest current Mac that runs it and cheapest at 15+ tok/s, $899 new (Apple list price, checked 2026-11-14), ~41 tok/s (estimated) once the GPU memory limit is raised. For Ollama: the gpt-oss:20b tag is about 14 GB, so it needs a 24 GB Mac; on 16 GB use a ~4-bit build after raising the GPU limit.

As an Amazon Associate, LLMCheck earns from qualifying purchases. Affiliate links cost you nothing extra and never influence the verdicts.

FAQ: GPT-oss 20B on a Mac

Can I run GPT-oss 20B on a Mac?

Yes. GPT-oss 20B needs about 13 GB of unified memory at a short (4K-token) context, so it runs on 53 of the 54 Mac configurations in the LLMCheck index; 43 of them hold it inside macOS’s default GPU memory limit and 10 only after raising that limit. The smallest Mac Apple sells new that runs it is the Mac mini M6 (16 GB).

How much RAM does GPT-oss 20B need?

About 13 GB at a 4K context: the weights plus about 2 GB for the runtime. Context adds KV cache: about 14 GB in total at 32K and 15 GB at 64K with the default 16-bit cache (about 14 GB at 64K with a q8 cache). These are estimates from the model’s config.json, not measurements.

How fast does GPT-oss 20B run on a Mac?

On an M5 Max, the chip the leaderboard quotes it on, the index figure is 141 tok/s (community run) at MXFP4. The best measured figure is 184 tok/s on an M5 Ultra 256 GB (community run, MXFP4). Every Mac’s figure is in the table on this page with its provenance: published, community run or estimated.

What is the cheapest Mac that runs GPT-oss 20B well?

Among Macs Apple sells new, the Mac mini M6 (16 GB), $899 new (Apple list price, checked 2026-11-14), is the cheapest that reaches 15 tok/s with GPT-oss 20B, at ~41 tok/s (estimated), after raising the GPU memory limit.

Do I need to raise the GPU memory limit to run GPT-oss 20B?

On 10 of the 54 configurations, yes: iMac M6 16 GB, Mac mini M6 16 GB, MacBook Air M5 16 GB and 7 more. macOS lets the GPU use only part of unified memory by default (about 12 GB on the iMac M6 (16 GB)), and GPT-oss 20B needs about 13 GB. Running sudo sysctl iogpu.wired_limit_mb=12288 in Terminal raises the limit on a 16 GB Mac until the next restart. From 24 GB up, every Mac Apple sells new holds it inside the default limit.

Which quantization of GPT-oss 20B should I use?

GPT-oss 20B ships natively in MXFP4 (about 11 GB of weights); re-quantizing it gains little. The index’s reference figure (141 tok/s on an M5 Max) is for its MXFP4 build.

How much context can GPT-oss 20B use on a Mac?

Its own window is 128K tokens. At macOS’s default GPU limit, the Mac mini M6 (24 GB) holds about 128K (the full window).

Cite this page

LLMCheck. (2026). Can I run GPT-oss 20B on a Mac? llmcheck.net. https://llmcheck.net/can-i-run/gpt-oss-20b/ (updated ). CC BY 4.0.

How to cite LLMCheck →Data: CC BY 4.0Markdown version

Related

Figures from the LLMCheck index, data version 2026-11-18. Verdicts and context limits are estimates from each model’s catalog memory figure and config.json.