gpt-oss-120b hardware requirements
What is gpt-oss-120b?
gpt-oss-120b is the larger of the two open-weight reasoning models OpenAI released in August 2025. Its model card aims it at general-purpose, high-reasoning work, with adjustable reasoning effort. It is trained to call functions, browse the web and run Python code, but the app running it has to supply those tools. It is a mixture-of-experts model with 5.1B of its 117B parameters active per token, released under the Apache 2.0 license.
What to know before running gpt-oss-120b
gpt-oss-120b comes only as an MXFP4 file of 63.4 GB, and no single consumer graphics card holds it. With CanRun's default of 32GB of system RAM it does not load on any of them, not even the 32GB GeForce RTX 5090. It is a mixture-of-experts model that uses only about 5B parameters per token, though, so with 64GB of system RAM the experts can sit in RAM: the GeForce RTX 3090 and the GeForce RTX 5060 Ti 16GB then run it slowly, by a theoretical estimate (±30%). An 8GB card such as the GeForce RTX 4060 needs 96GB of system RAM to load it.
It fits completely only with 96GB or more of unified memory. A 64GB Mac or a 64GB DGX Spark falls short: the model needs 64.3 GB, more than the GPU may use out of 64GB. The 96GB Apple M3 Ultra runs it great, and so do the 128GB Ryzen AI Max+ 395 (Strix Halo), NVIDIA DGX Spark and Apple M4 Max (40-core GPU). Like gpt-oss-20b, it always reasons before answering, so CanRun holds it to the 1.5× stricter speed bar for thinking models.
About the model
- Parameters
- 116.8B total · 5.1B active (MoE)
- Attention
- Sliding-window attention
- Max context
- 131,072 tokens
- Released
- 2025-08-05
- Model card
- gpt-oss-120b on Hugging Face
Other sizes in this series
| Model | Parameters | Baseline quant | Memory needed (8k) | GeForce RTX 4060 | GeForce RTX 5060 Ti 16GB | GeForce RTX 3090 |
|---|---|---|---|---|---|---|
| gpt-oss-20b | 20.9B total · 3.6B active (MoE) | MXFP4 | 12.3 GB | Runs well | Runs great | Runs great |
| gpt-oss-120bthis model | 116.8B total · 5.1B active (MoE) | MXFP4 | 63.6 GB | Won't run | Won't run | Won't run |
GGUF files
| Quant | File size | Quality | Published by |
|---|---|---|---|
| MXFP4baseline | 63.39 GB | Near-lossless | llama.cpp team · ggml-org |
Memory needed by quant and context
| Quant | 4k | 8k | 16k | 32k | 64k | 128k |
|---|---|---|---|---|---|---|
| MXFP4 | 63.4 | 63.6 | 64.0 | 64.8 | 66.3 | 69.3 |
Which GPUs can run gpt-oss-120b?
How to read the verdicts
- Runs great
- Fully on the GPU at 20 tok/s or more
- Runs well
- Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
- Runs slowly
- 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
- Won't run
- Does not fit, or under 2 tok/s
Filter this list by budget, type and context in the calculator
Scroll sideways to see every column.
| Hardware | Verdict | Speed | Quant | Memory | Runs as | Context | Notes | Price |
|---|---|---|---|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell | Runs great | est. 237.1 tok/s189.7–284.5calibrated estimate ±20% | MXFP4 | 64.2 / 96.0 GB | Full GPU | up to 128k | — | — |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | Runs great | est. 237.1 tok/s189.7–284.5calibrated estimate ±20% | MXFP4 | 64.2 / 96.0 GB | Full GPU | up to 128k | — | — |
| Apple M5 Ultra · 96 GB | Runs great | est. 131.0 tok/s104.8–157.2calibrated estimate ±20% | MXFP4 | 64.3 / 67.2 GB | Unified memory | up to 64k | — | — |
| Apple M5 Max (40-core GPU) · 128 GB | Runs great | est. 102.4 tok/s81.9–122.8calibrated estimate ±20% | MXFP4 | 64.3 / 89.6 GB | Unified memory | up to 128k | — | — |
| Apple M3 Ultra · 96 GB | Runs great | est. 96.0 tok/s76.8–115.2calibrated estimate ±20% | MXFP4 | 64.3 / 67.2 GB | Unified memory | up to 64k | — | — |
| Apple M4 Max (40-core GPU) · 128 GB | Runs great | est. 92.5 tok/s74.0–111.0calibrated estimate ±20% | MXFP4 | 64.3 / 89.6 GB | Unified memory | up to 128k | — | — |
| Apple M2 Ultra · 128 GB | Runs great | 79.7 tok/s8k estimate 94.2 tok/s (75.4–113.1, calibrated estimate ±20%)measured (1 run, 2k context) | MXFP4 | 64.3 / 89.6 GB | Unified memory | up to 128k | — | — |
| NVIDIA DGX Spark · 128 GB | Runs great | 58.7 tok/s8k estimate 50.8 tok/s (40.6–60.9, calibrated estimate ±20%)measured (1 run, 2k context) | MXFP4 | 64.3 / 116.9 GB | Unified memory | up to 64k | — | $3,999 launch MSRP (128 GB) |
| Ryzen AI Max+ 395 (Strix Halo) · 128 GB | Runs great | 50.2 tok/s8k estimate 48.5 tok/s (38.8–58.1, calibrated estimate ±20%)measured (1 run, 512 context) | MXFP4 | 64.3 / 89.6 GB | Unified memory | up to 64k | — | $1,999 launch MSRP (128 GB) |
| DDR5-6000 dual-channel CPU · 96 GB | Runs slowly | est. 13.4 tok/s10.8–16.1calibrated estimate ±20% | MXFP4 | 63.7 GB RAM | CPU only | up to 128k |
| — |
| DDR5-5600 dual-channel CPU · 96 GB | Runs slowly | est. 12.6 tok/s10.0–15.1calibrated estimate ±20% | MXFP4 | 63.7 GB RAM | CPU only | up to 128k |
| — |
| Apple M4 · 32 GB | Won't rungpt-oss-120b verdict on the Apple M4 | — | MXFP4 | needs 64.3 GB | — | — |
| — |
| Apple M4 Max (32-core GPU) · 36 GB | Won't run | — | MXFP4 | needs 64.3 GB | — | — |
| — |
| Apple M4 Pro · 64 GB | — | MXFP4 | needs 64.3 GB | — | — |
| — | |
| Apple M5 · 32 GB | Won't run | — | MXFP4 | needs 64.3 GB | — | — |
| — |
| Apple M5 Max (32-core GPU) · 36 GB | Won't run | — | MXFP4 | needs 64.3 GB | — | — |
| — |
| Apple M5 Pro · 64 GB | Won't run | — | MXFP4 | needs 64.3 GB | — | — |
| — |
| Apple M6 · 32 GB | Won't run | — | MXFP4 | needs 64.3 GB | — | — |
| — |
| Intel Arc B580 | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $290 street (as of 2026-08-09) |
| DDR4-3200 dual-channel CPU · 64 GB | Won't run | — | MXFP4 | needs 63.7 GB | — | — |
| — |
| GeForce GTX 1080 Ti | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $699 launch MSRP |
| GeForce RTX 2080 Ti | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $999 launch MSRP |
| GeForce RTX 3050 8GB | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $249 launch MSRP |
| GeForce RTX 3060 12GB | — | MXFP4 | needs 63.6 GB | — | — |
| $250 street (as of 2026-08-09) | |
| GeForce RTX 3070 | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $499 launch MSRP |
| GeForce RTX 3080 10GB | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $699 launch MSRP |
| GeForce RTX 3080 12GB | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $799 launch MSRP |
| GeForce RTX 3090 | — | MXFP4 | needs 63.6 GB | — | — |
| $1,050 street (as of 2026-08-09) | |
| GeForce RTX 3090 Ti | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $1,999 launch MSRP |
| GeForce RTX 4060 | — | MXFP4 | needs 63.6 GB | — | — |
| $299 launch MSRP | |
| GeForce RTX 4060 Laptop | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| — |
| GeForce RTX 4060 Ti 16GB | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $499 launch MSRP |
| GeForce RTX 4060 Ti 8GB | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $399 launch MSRP |
| GeForce RTX 4070 | — | MXFP4 | needs 63.6 GB | — | — |
| $599 launch MSRP | |
| GeForce RTX 4070 Laptop | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| — |
| GeForce RTX 4070 Super | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $599 launch MSRP |
| GeForce RTX 4070 Ti Super | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $799 launch MSRP |
| GeForce RTX 4080 | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $1,199 launch MSRP |
| GeForce RTX 4080 Laptop | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| — |
| GeForce RTX 4080 Super | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $999 launch MSRP |
| GeForce RTX 4090 | — | MXFP4 | needs 63.6 GB | — | — |
| $2,755 street (as of 2026-08-09) | |
| GeForce RTX 4090 Laptop | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| — |
| GeForce RTX 5060 | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $339 street (as of 2026-08-09) |
| GeForce RTX 5060 Ti 16GB | — | MXFP4 | needs 63.6 GB | — | — |
| $569 street (as of 2026-08-09) | |
| GeForce RTX 5060 Ti 8GB | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $429 street (as of 2026-08-09) |
| GeForce RTX 5070 | — | MXFP4 | needs 63.6 GB | — | — |
| $629 street (as of 2026-08-09) | |
| GeForce RTX 5070 Ti | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $949 street (as of 2026-08-09) |
| GeForce RTX 5070 Ti Laptop | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| — |
| GeForce RTX 5080 | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $1,256 street (as of 2026-08-09) |
| GeForce RTX 5080 Laptop | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| — |
| GeForce RTX 5090 | — | MXFP4 | needs 63.6 GB | — | — |
| $4,200 street (as of 2026-08-09) | |
| GeForce RTX 5090 Laptop | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| — |
| Radeon RX 7800 XT | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $499 launch MSRP |
| Radeon RX 7900 XT | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $899 launch MSRP |
| Radeon RX 7900 XTX | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $999 launch MSRP |
| Radeon RX 9060 XT 16GB | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $449 street (as of 2026-08-09) |
| Radeon RX 9070 | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $639 street (as of 2026-08-09) |
| Radeon RX 9070 XT | Won't run | — | MXFP4 | needs 63.6 GB | — | — |
| $689 street (as of 2026-08-09) |
Models with a thinking mode spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s). With thinking turned off, a model may do one grade better.
Cheapest GPUs that run it
No GPU with a current street price reaches Runs great or Runs well at MXFP4. The table above lists Macs, unified-memory PCs and CPU setups.
Measured results for gpt-oss-120b
Public benchmarks we calibrate against. Their conditions (context, backend, flags) can differ from the estimates above.
| Hardware | Quant | Backend | Context | Prompt (tok/s) | Generation (tok/s) | Flags | Source | Measured |
|---|---|---|---|---|---|---|---|---|
| Apple M2 Ultra | MXFP4 | llama.cpp | 2k | 1,245.0 | 79.7 | — | github.com | 2025-08-18 |
| Ryzen AI Max+ 395 (Strix Halo) | MXFP4 | llama.cpp | 512 | 707.0 | 50.2 | ROCm 7.0.2, no mmap | quozul.dev | 2025-10-27 |
| NVIDIA DGX Spark | MXFP4 | llama.cpp | 2k | 2,444.0 | 58.7 | CUDA, no mmap | github.com | 2026-02-05 |
| GeForce RTX 5090 | MXFP4 | llama.cpp | 4k | — | 9.6 | --n-cpu-moe 21, DDR4 server | hardware-corner.net | 2026-08-09 |
Frequently asked questions
How much VRAM does gpt-oss-120b need?
At MXFP4 the file is 63.39 GB; with 8k context a graphics card needs about 63.6 GB (weights 62.76 GB, KV cache 0.3 GB, compute buffer 0.6 GB). Part of the file stays off the graphics card: llama.cpp keeps the input embedding table in system RAM. A Mac counts the whole file, so it needs a little more.
What is the cheapest GPU that runs gpt-oss-120b well?
No GPU with a current street price reaches Runs great or Runs well at MXFP4. Counting hardware with only a launch price too, the best option is NVIDIA DGX Spark · 128 GB: Runs great.
Can gpt-oss-120b run on an 8 GB, 16 GB or 24 GB GPU?
At MXFP4 with 8k context: GeForce RTX 4060 — Won't run; GeForce RTX 5060 Ti 16GB — Won't run; GeForce RTX 3090 — Won't run.
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.