Can a Mac run gpt-oss 20B?
Yes. gpt-oss 20B needs about 24 GB even for short chats and the same 24 GB for long 32,000-token work, by our cautious rule, not measured by us.
On the MacBook Pro M1 Pro, expect about 21 to 42 tokens/s, estimated. On the Mac Studio M1 Ultra, expect about 83 to 167 tokens/s, estimated.
Size and tasks
This download lets you chat and help with code. It takes 11.18 GB of disk space. At 21B MoE / 3.6B active, it sits alongside 15 smaller models and 17 larger models in our comparison by total parameter count. MoE means mixture of experts, with the active part used for each token.
This is the smaller gpt-oss download in our list, with support for chat and coding. It uses a mixture of experts, so the active part is smaller than the whole model. Allow disk space for the complete file when planning the download. Its memory requirement rules out some lower Mac configurations, while others offer a larger option that fits. Nearby dense models give you different file sizes and supported work to compare. We have no measured rate for this model. The estimates are a rough guide to the pace on each chip.
The exact model ID is gpt-oss-20b.
How we work this out
| Calculation | Short chats | Long work |
|---|---|---|
| Tokens | 8,192 | 32,768 |
| KV sum in GB | 0.402653184 | 1.610612736 |
| KV rounded to two decimals in GB | 0.4 | 1.61 |
| Need in GB | 13.58 | 14.79 |
| First memory size in GB | 24 | 24 |
| Usable GPU memory in GB, rounded | 16 | 16 |
| Close threshold in GB, rounded | 15.2 | 15.2 |
| Close margin step | Keep this size | Keep this size |
| Minimum Mac memory in GB | 24 | 24 |
| Usable GPU memory is 2/3 of Mac memory up to 36 GB and 3/4 above that. Pick the first size that holds the need from 8, 16, 18, 24, 32, 36, 48, 64, 96, 128, 192, 256, 512 GB. If need is above 95% of usable memory, within 5%, take the next memory size. | ||
OpenAI's memory figure
“the gpt-oss-20b model run within 16GB of memory”
OpenAI, model card, read 11 November 2026.
On a Mac, macOS lets the GPU use about two thirds of memory by default. A 16 GB Mac has about 10.7 GB for the model, less than the 11.18 GB of 4-bit weights this page's file takes. OpenAI's 16 GB figure is for its own MXFP4 weights on a GPU. Our 24 GB short-chat figure comes from our cautious rule, with a 16-bit cache and every attention layer counted in full.
Which Macs can run it
The memory checks, comparisons and upgrade advice below use our long 32,000-token work figure. Use the short-chat figure above for shorter conversations.
Your chip name alone cannot settle the fit. Match the memory in your Mac to the options below. Across 38 Mac models, 33 offer enough memory in at least one configuration. That leaves five Macs needing a larger capacity than their listed options offer. Read tight as fitting with little memory to spare.
We have estimated every chip rate for this download. The memory minimum comes from a rule and has not been measured by us. Open the speed details to see how a range was worked out.
| Mac and chip | Every listed memory option | Chip speed source |
|---|---|---|
| First fitting configuration 24 GB | ||
| MacBook Air M2 Apple M2 | 8 GB does not fit. Next 24 GB configuration. 16 GB does not fit. Next 24 GB configuration. 24 GB tight. |
about 10 to 21 tokens/s
estimate, not measured by us
How we estimate speed10.43 to 20.88 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Pro 13 M2 Apple M2 | 8 GB does not fit. Next 24 GB configuration. 16 GB does not fit. Next 24 GB configuration. 24 GB tight. |
about 10 to 21 tokens/s
estimate, not measured by us
How we estimate speed10.43 to 20.88 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| Mac mini M2 Apple M2 | 8 GB does not fit. Next 24 GB configuration. 16 GB does not fit. Next 24 GB configuration. 24 GB tight. |
about 10 to 21 tokens/s
estimate, not measured by us
How we estimate speed10.43 to 20.88 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Air 13 M3 Apple M3 | 8 GB does not fit. Next 24 GB configuration. 16 GB does not fit. Next 24 GB configuration. 24 GB tight. |
about 10 to 21 tokens/s
estimate, not measured by us
How we estimate speed10.43 to 20.88 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Air 15 M3 Apple M3 | 8 GB does not fit. Next 24 GB configuration. 16 GB does not fit. Next 24 GB configuration. 24 GB tight. |
about 10 to 21 tokens/s
estimate, not measured by us
How we estimate speed10.43 to 20.88 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| iMac M3 Apple M3 | 8 GB does not fit. Next 24 GB configuration. 16 GB does not fit. Next 24 GB configuration. 24 GB tight. |
about 10 to 21 tokens/s
estimate, not measured by us
How we estimate speed10.43 to 20.88 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Pro 14 M4 Apple M4 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 13 to 25 tokens/s
estimate, not measured by us
How we estimate speed12.52 to 25.05 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| iMac M4 Apple M4 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 13 to 25 tokens/s
estimate, not measured by us
How we estimate speed12.52 to 25.05 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| Mac mini M4 Apple M4 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 13 to 25 tokens/s
estimate, not measured by us
How we estimate speed12.52 to 25.05 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| Mac mini M4 Pro Apple M4 Pro | 24 GB tight. 48 GB fits. 64 GB fits. |
about 28 to 57 tokens/s
estimate, not measured by us
How we estimate speed28.48 to 56.98 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 273 GB/s divided by active weights in GB |
| MacBook Pro M4 Pro Apple M4 Pro | 24 GB tight. 48 GB fits. |
about 28 to 57 tokens/s
estimate, not measured by us
How we estimate speed28.48 to 56.98 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 273 GB/s divided by active weights in GB |
| MacBook Air 13 M4 Apple M4 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 13 to 25 tokens/s
estimate, not measured by us
How we estimate speed12.52 to 25.05 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| MacBook Air 15 M4 Apple M4 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 13 to 25 tokens/s
estimate, not measured by us
How we estimate speed12.52 to 25.05 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| MacBook Air 13 M5 Apple M5 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 16 to 32 tokens/s
estimate, not measured by us
How we estimate speed15.96 to 31.94 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153 GB/s divided by active weights in GB |
| MacBook Air 15 M5 Apple M5 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 16 to 32 tokens/s
estimate, not measured by us
How we estimate speed15.96 to 31.94 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153 GB/s divided by active weights in GB |
| MacBook Pro 14 M6 Apple M6 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 16 to 35 tokens/s
estimate, not measured by us
How we estimate speed15.96 to 35.49 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
| MacBook Pro M5 Pro Apple M5 Pro | 24 GB tight. 48 GB fits. 64 GB fits. |
about 32 to 64 tokens/s
estimate, not measured by us
How we estimate speed32.03 to 64.08 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 307 GB/s divided by active weights in GB |
| Mac mini M6 Apple M6 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 16 to 35 tokens/s
estimate, not measured by us
How we estimate speed15.96 to 35.49 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
| Mac mini M5 Pro Apple M5 Pro | 24 GB tight. 48 GB fits. 64 GB fits. |
about 32 to 64 tokens/s
estimate, not measured by us
How we estimate speed32.03 to 64.08 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 307 GB/s divided by active weights in GB |
| iMac M6 Apple M6 | 16 GB does not fit. Next 24 GB configuration. 24 GB tight. 32 GB fits. |
about 16 to 35 tokens/s
estimate, not measured by us
How we estimate speed15.96 to 35.49 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
| First fitting configuration 32 GB | ||
| MacBook Pro M1 Pro Apple M1 Pro | 16 GB does not fit. Next 32 GB configuration. 32 GB fits. |
about 21 to 42 tokens/s
estimate, not measured by us
How we estimate speed20.87 to 41.75 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 200 GB/s divided by active weights in GB |
| MacBook Pro M1 Max Apple M1 Max | 32 GB fits. 64 GB fits. |
about 42 to 83 tokens/s
estimate, not measured by us
How we estimate speed41.74 to 83.49 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 400 GB/s divided by active weights in GB |
| MacBook Pro M2 Pro Apple M2 Pro | 16 GB does not fit. Next 32 GB configuration. 32 GB fits. |
about 21 to 42 tokens/s
estimate, not measured by us
How we estimate speed20.87 to 41.75 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 200 GB/s divided by active weights in GB |
| MacBook Pro M2 Max Apple M2 Max | 32 GB fits. 64 GB fits. 96 GB fits. |
about 42 to 83 tokens/s
estimate, not measured by us
How we estimate speed41.74 to 83.49 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 400 GB/s divided by active weights in GB |
| First fitting configuration 36 GB | ||
| MacBook Pro M3 Pro Apple M3 Pro | 18 GB does not fit. Next 36 GB configuration. 36 GB fits. |
about 16 to 31 tokens/s
estimate, not measured by us
How we estimate speed15.65 to 31.31 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 150 GB/s divided by active weights in GB |
| MacBook Pro M3 Max Apple M3 Max | 36 GB fits. 48 GB fits. 64 GB fits. 96 GB fits. 128 GB fits. |
about 31 to 83 tokens/s
estimate, not measured by us
How we estimate speed31.3 to 83.49 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 300-400 GB/s divided by active weights in GB |
| MacBook Pro M4 Max Apple M4 Max | 36 GB fits. 48 GB fits. 64 GB fits. 128 GB fits. |
about 43 to 114 tokens/s
estimate, not measured by us
How we estimate speed42.78 to 113.96 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 410-546 GB/s divided by active weights in GB |
| MacBook Pro M5 Max Apple M5 Max | 36 GB fits. 48 GB fits. 64 GB fits. 128 GB fits. |
about 48 to 128 tokens/s
estimate, not measured by us
How we estimate speed48 to 128.15 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 460-614 GB/s divided by active weights in GB |
| Mac Studio M5 Max Apple M5 Max | 36 GB fits. 48 GB fits. 64 GB fits. 128 GB fits. |
about 48 to 128 tokens/s
estimate, not measured by us
How we estimate speed48 to 128.15 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 460-614 GB/s divided by active weights in GB |
| First fitting configuration 64 GB | ||
| Mac Studio M1 Ultra Apple M1 Ultra | 64 GB fits. 128 GB fits. |
about 83 to 167 tokens/s
estimate, not measured by us
How we estimate speed83.48 to 166.97 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 800 GB/s divided by active weights in GB |
| Mac Studio M2 Ultra Apple M2 Ultra | 64 GB fits. 128 GB fits. 192 GB fits. |
about 83 to 167 tokens/s
estimate, not measured by us
How we estimate speed83.48 to 166.97 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 800 GB/s divided by active weights in GB |
| First fitting configuration 96 GB | ||
| Mac Studio M3 Ultra Apple M3 Ultra | 96 GB fits. 256 GB fits. 512 GB fits. |
about 85 to 171 tokens/s
estimate, not measured by us
How we estimate speed85.46 to 170.94 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 819 GB/s divided by active weights in GB |
| Mac Studio M5 Ultra Apple M5 Ultra | 96 GB fits. 256 GB fits. 512 GB fits. |
about 125 to 250 tokens/s
estimate, not measured by us
How we estimate speed125.22 to 250.45 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 1200 GB/s divided by active weights in GB |
| No fitting configuration listed | ||
| MacBook Air M1 Apple M1 | 8 GB does not fit. Next 24 GB on a different listed Mac. 16 GB does not fit. Next 24 GB on a different listed Mac. |
about 7.1 to 14 tokens/s
estimate, not measured by us
How we estimate speed7.12 to 14.25 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| MacBook Pro 13 M1 Apple M1 | 8 GB does not fit. Next 24 GB on a different listed Mac. 16 GB does not fit. Next 24 GB on a different listed Mac. |
about 7.1 to 14 tokens/s
estimate, not measured by us
How we estimate speed7.12 to 14.25 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| Mac mini M1 Apple M1 | 8 GB does not fit. Next 24 GB on a different listed Mac. 16 GB does not fit. Next 24 GB on a different listed Mac. |
about 7.1 to 14 tokens/s
estimate, not measured by us
How we estimate speed7.12 to 14.25 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| iMac M1 Apple M1 | 8 GB does not fit. Next 24 GB on a different listed Mac. 16 GB does not fit. Next 24 GB on a different listed Mac. |
about 7.1 to 14 tokens/s
estimate, not measured by us
How we estimate speed7.12 to 14.25 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| MacBook Neo Apple A18 Pro | 8 GB does not fit. Next 24 GB on a different listed Mac. |
about 6.3 to 13 tokens/s
estimate, not measured by us
How we estimate speed6.26 to 12.53 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 60 GB/s divided by active weights in GB |
Nearby models by total parameter count
Compare these similar-sized downloads if you want a different memory requirement or set of capabilities.
To chat and read images, you could try Mistral Small 3.2 24B Instruct 2506. Its long-work minimum is 36 GB of memory, 12 GB more. Its size is 24B dense. Its download uses 2.08 GB more disk space. Switching means giving up coding help.
Devstral Small 2 24B Instruct 2512 can chat, help with code, and read images. Its size is 24B dense. Its download uses 3.92 GB more disk space. Its long-work minimum is 36 GB of memory, 12 GB more.
You could use Gemma 4 26B-A4B to chat. Its measured minimum is 16 GB of memory, 8 GB less. Its download uses 4.43 GB more disk space. Its size is 26B MoE / 4B active. Switching means giving up coding help.
See the 24 GB memory page to compare the other models you could run with that much memory.
Context and data dates
Memory minimums use a 32K context; longer contexts need more memory. Conversation length matters to memory use. The default is 32768 tokens of context, meaning the text the model keeps in view. Keeping a longer exchange available asks more of your Mac's memory.
The model source date is 2026-11-08. Figures as of 8 November 2026.
Mac source.
Apple technical specifications (apple.com/mac and the specs page of each Mac, read 2026-11-07; the iMac M6 page says only "up to 170GB/s", so its low end, 153 GB/s, is the 16 GB figure for the same M6 chip on the Mac mini page); table last changed 2026-11-07Chip ratio source.
llama.cpp's public Apple-silicon benchmark (github.com/ggml-org/llama.cpp/discussions/4167, read 2026-11-07): LLaMA 7B Q4_0 text generation, as a ratio to the M1 Ultra 48-core GPUModel date source.
Dates come from each source text; otherwise table last changed 2026-10-19Outlier Pro: $39 once · Black Friday sale: $16.99 once through December 2. Then $39.
Free to start (Nano and Lite). macOS 15+, Apple silicon. Refund window: 30 days, no questions asked. Priced in US dollars. Checkout shows your currency and any local tax.
Try Outlier on your Mac
Choose a Mac configuration with enough memory for gpt-oss 20B, then download Outlier to get started.
Download Outlier