How to Deploy gpt-oss 120B Locally
gpt-oss 120B is a MoE model with 116.8B total parameters and 5.1B active per token. At Q4 it needs 67GB of VRAM to sit entirely on the GPU; with experts offloaded to system RAM it needs only 4GB of VRAM plus 65GB of RAM. The cheapest card that runs it today is the Ryzen AI Max+ 395 128GB at about $3,913, at roughly 34.6 tokens/s. In offload mode the cheapest card that runs it is the Tesla V100 16GB, at roughly 21 tokens/s.
gpt-oss 120B details
How much VRAM does gpt-oss 120B need for local deployment?
MoE: all 116.8B parameters must stay resident in memory, while speed is set by the 5.1B active per token.
| Quantization | WeightsGB | Total at 8KGB | Total at 32KGB | Total at 128KGB |
|---|---|---|---|---|
| 1-bit | 27.4est. | 29 | 30.7 | 37.4 |
| 2-bit | 43.4est. | 45 | 46.7 | 53.4 |
| 3-bit | 62.6GGUF | 57.5 | 59.2 | 66 |
| 4-bit | 62.8GGUF | 66.7 | 68.4 | 75.1 |
| 5-bit | 62.9GGUF | 80.4 | 82.1 | 88.8 |
| 8-bit | 63.4GGUF | 122.7 | 124.4 | 131.1 |
| FP16 | 65.4GGUF | 230.1 | 231.7 | 238.5 |
Total = weights + KV cache (FP16) + 1GB runtime overhead. Weights marked GGUF are measured file sizes from the repository; the rest are estimated from bytes per parameter. Contexts beyond this model's 128K limit show a dash. Each tier takes whichever file of that bit width the repository has, preferring unsloth dynamic quants, so the exact file name varies by model.
Which GPUs can run gpt-oss 120B?
AmazonJP
| Device | VRAMGB | Price | Est. t/s | Suggested context |
|---|---|---|---|---|
 Ryzen AI Max+ 395 128GB | 128 | $3,9136 listings | 34.6 | 128K |
 DGX Spark 128GB | 128 | $8,2233 listings | 48.3 | 128K |
| RTX PRO 5000 Blackwell 72GB | 72 | —No listings | 135.3 | 32K |
| RTX PRO 6000D | 84 | —No listings | 144.8 | 128K |
| RTX PRO 6000 Blackwell | 96 | —No listings | 152.9 | 128K |
| Mac Studio M4 Max 128GB | 128 | —No listings | 68.4 | 128K |
| Mac Studio M5 Max 128GB | 128 | —No listings | 74.4 | 128K |
| MacBook Pro M4 Max 128GB | 128 | —No listings | 68.4 | 128K |
| MacBook Pro M5 Max 128GB | 128 | —No listings | 74.4 | 128K |
| Mac Studio M3 Ultra 512GB | 512 | —No listings | 90.3 | 128K |
Fits when weights + KV cache + runtime overhead is at or below usable memory, single card. "Offload" on MoE models means it runs with expert weights in system RAM (only attention layers and KV cache stay in VRAM); that speed assumes 70 GB/s RAM bandwidth. The context column is the largest tier that fits.
Recommended GPUs for gpt-oss 120B with CPU/GPU offloading
AmazonJP
Expert weights live in system RAM (needs at least 65 GB); VRAM holds only attention layers, shared experts and the KV cache, at least 4 GB at Q4. Speed is bound by RAM bandwidth (estimated at 70 GB/s) and is far slower than a full-VRAM setup.
Deploy gpt-oss 120B locally with Ollama, llama.cpp or vLLM
Ollama pulls and runs it in one command; the quantization comes from the official tag.
ollama run gpt-oss:120b
llama.cpp pulls the GGUF straight from Hugging Face; -c sets the context length.
llama-server -hf unsloth/gpt-oss-120b-GGUF:Q4_K_M -c 8192
vLLM serves the original-precision weights, which needs far more memory than GGUF and usually more than one GPU.
vllm serve openai/gpt-oss-120b
File names follow that week's model-library snapshot; the definitions are in the methodology. Methodology
FAQ
How much VRAM does gpt-oss 120B need?
At Q4_K_M with an 8K context gpt-oss 120B needs about 67GB of VRAM; 84GB leaves comfortable headroom.
What is the cheapest GPU that runs gpt-oss 120B?
The Ryzen AI Max+ 395 128GB: 128GB of VRAM, currently about $3,913, at roughly 34.6 tokens/s.
Can you run gpt-oss 120B with Ollama?
Yes: ollama run gpt-oss:120b.