Ollama
Ollama
Price: Free (MIT-licensed); Ollama also offers cloud-hosted models
- macOS 14 Sonoma or newer
- Windows 10 22H2 or newer
- Linux
- Docker
Ollama on GitHub · company-reported
The short answer
How much memory it needs.
Its usual download, the MXFP4 file, is 65 GB. It runs comfortably with about 87 GB of memory or more, and is a tight fit from 65 GB up to 87 GB.
Our rule of thumb, not the makers': a file up to three quarters of the memory runs comfortably, leaving room for the system, other apps and the conversation. A tight fit means it fits, with little room left: close other apps and keep conversations short. Ollama says longer conversations need more memory. On a Mac, and on PCs with shared memory, the model uses the computer's main memory; with a graphics card, it runs fastest when the whole file fits in the card's own memory.
Which machines
Machines with enough memory.
For each machine we list, the smallest memory it can be ordered with that fits the 65 GB file, by the same rule. Worked out from sizes alone: no one ran gpt-oss 120b on each of these.
| Machine | Memory it comes with | For gpt-oss 120b | Starting price |
|---|---|---|---|
| MacBook Pro 14-inch (M5 Pro / M5 Max)laptop, shared memory | 24, 36, 48, 64, 128 GB | Runs comfortably with 128 GB | $2,199with 24 GB, maker's price, Mar 3, 2026 |
| MacBook Pro 16-inch (M5 Pro / M5 Max)laptop, shared memory | 24, 36, 48, 64, 128 GB | Runs comfortably with 128 GB | $2,699with 24 GB, maker's price, Mar 3, 2026 |
| Mac Studio (M5 Max / M5 Ultra)desktop, shared memory | 36, 48, 64, 96, 128, 256, 512 GB | Runs comfortably with 96 GB | $2,499with 36 GB, maker's price, Aug 25, 2026 |
| NVIDIA DGX Sparkmini PC, shared memory | 64, 128 GB | Runs comfortably with 128 GB | $4,999with 64 GB, maker's price, Oct 20, 2026 |
| Framework Desktop (AMD Ryzen AI Max / Max+)desktop, shared memory | 32, 64, 128 GB | Runs comfortably with 128 GB | $1,269with 32 GB, maker's price, Oct 20, 2026 |
| Mac mini (M6)desktop, shared memory | 16, 24, 32 GB | Too big, even at 32 GB | $899with 16 GB, maker's price, Aug 25, 2026 |
| Mac mini (M5 Pro)desktop, shared memory | 24, 48, 64 GB | Too big, even at 64 GB | $1,699with 24 GB, maker's price, Aug 25, 2026 |
| MacBook Air 13-inch (M5)laptop, shared memory | 16, 24, 32 GB | Too big, even at 32 GB | $1,099with 16 GB, maker's price, Mar 3, 2026 |
| MacBook Air 15-inch (M5)laptop, shared memory | 16, 24, 32 GB | Too big, even at 32 GB | $1,299with 16 GB, maker's price, Mar 3, 2026 |
| MacBook Pro 14-inch (M5)laptop, shared memory | 16, 24, 32 GB | Too big, even at 32 GB | $1,599with 16 GB, maker's price, Oct 15, 2025 |
| GeForce RTX 5090graphics card, memory on the card | 32 GB | Too big, even at 32 GB | $1,999with 32 GB, launch price, Jan 6, 2025 |
| GeForce RTX 5080graphics card, memory on the card | 16 GB | Too big, even at 16 GB | $999with 16 GB, launch price, Jan 6, 2025 |
| GeForce RTX 5070 Tigraphics card, memory on the card | 16 GB | Too big, even at 16 GB | $749with 16 GB, launch price, Jan 6, 2025 |
| GeForce RTX 5070graphics card, memory on the card | 12 GB | Too big, even at 12 GB | $549with 12 GB, launch price, Jan 6, 2025 |
| GeForce RTX 5060 Tigraphics card, memory on the card | 8, 16 GB | Too big, even at 16 GB | $379with 8 GB, launch price, Apr 15, 2025 |
| GeForce RTX 5060graphics card, memory on the card | 8 GB | Too big, even at 8 GB | $299with 8 GB, launch price, Apr 15, 2025 |
The starting price buys the smallest memory; more memory costs more, and makers' stores show those prices only as you configure. Graphics card prices are NVIDIA's launch prices from 2025, and a card also needs a PC to go in. Copilot+ PCs are left out: Microsoft's pages give no memory minimum. All machines and prices
The downloads
Every file size, and where to get it.
Listed on Ollama's library and LM Studio, read Oct 20, 2026. Fewer bits per number makes a smaller file. The usual home choice, its smallest 4-bit file, is the one we size it by.
Download sizes
| File | Size | Comfortable from |
|---|---|---|
| MXFP4gpt-oss:120b, Oct 20, 2026 | 65 GBusual | 87 GB |
| MXFP4openai/gpt-oss-120b, Oct 20, 2026 | 65 GB | 87 GB |
The model
- Made by
- OpenAIBased in US
- Family
- gpt-oss
- Size
- 117B parameters, 5.1B activeOnly part of the model works on each word, but the whole file still has to fit in memory
- Reads at once
- 128K tokensLonger conversations take more memory
- License
- Apache 2.0
- Online version
- gpt-oss-120b: price and rankings
“running efficiently on a single 80 GB GPU”
Our note: LM Studio: 'The largest one may require up to 65 GB.' Released August 2025.
How fast it runs
Speed, measured.
Tokens a second as it writes, from a published benchmark that ran this model. A token is a word or part of a word.
The full setup is in the source it links to. Other machines will differ. How chips compare on a small model
In the makers' words
What the apps say about memory.
“Setting a larger context length will increase the amount of memory required to run a model.”
“16GB+ RAM recommended.”
“(8GB RAM for 3B models, 16GB for 7B, 32GB for 13B)”
“When using GPU inference new models must be able to completely fit in VRAM to allow concurrent model loads.”
The apps
Apps that run downloaded models.
Free apps that download open models and run them on your machine. gpt-oss 120b's files are listed on Ollama's library and LM Studio; check an app's own library for the model before you rely on it.
What each app doesLM Studio
LM Studio
Price: Free tier $0 (Run local LLMs on your machine); paid Bionic+ $20/month and Pro $100/month add cloud models
- macOS 14 or newer on Apple Silicon
- Windows x64
- Linux x64 and ARM64
LM Studio pricing page and system requirements · company-reported
llama.cpp
ggml-org (open-source project)
Price: Free
- macOS
- Windows
- Docker
- Android
- Pre-built binaries or build from source
llama.cpp on GitHub · company-reported
Jan
Jan (janhq)
Price: Free (Free & Open source)
- Windows 10 or later
- macOS 13.6 or later
- Linux
Jan on GitHub · company-reported
GPT4All
Nomic AI
Price: Free
- Windows x64 and ARM
- macOS 12.6 Monterey or newer
- Linux x86-64
GPT4All on GitHub · company-reported
Foundry Local
Microsoft
Price: There are no per-token costs and No Azure subscription is required.
- Windows
- macOS
- Linux
Microsoft Learn · company-reported
More of gpt-oss, and models of a similar size