How much VRAM does gpt-oss 120B need?
gpt-oss 120B needs 59.0 GB for its weights as released (MXFP4), 60.2 GB with an 8K context and 61.1 GB with 32K.
- Parameters
- 117B
- Active per token
- 5.1B
- Longest context
- 128K
- Released
- Aug 2025
- License
- Apache 2.0
- Strongest at
- Chat, Coding
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| As released (MXFP4) | 59.0 GB | 60.2 GB | 61.1 GB | 64.5 GB |
| 3-bit | 53.2 GB | 54.4 GB | 55.2 GB | 58.6 GB |
| 2-bit | 40.8 GB | 42.0 GB | 42.9 GB | 46.2 GB |
| 1.6-bit | 22.4 GB | 23.6 GB | 24.5 GB | 27.9 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | Too big | |||
| RTX 3090 | Too big | |||
| RTX 4090 | Too big | |||
| RTX 5090 | Too big | |||
| 2× RTX 3090 | 2-bit | 88K | 197 tok/s | 0.5 s |
| RTX PRO 6000 Blackwell | MXFP4 | 128K | 197 tok/s | 0.2 s |
| M4 Max 128 GB | MXFP4 | 128K | 65 tok/s | 4.6 s |
| M5 Max 128 GB | MXFP4 | 128K | 71 tok/s | 1.3 s |
| Ryzen AI Max+ 395 128 GB | MXFP4 | 128K | 37 tok/s | 2.7 s |
| DGX Spark 128 GB | MXFP4 | 128K | 47 tok/s | 1.3 s |
| M3 Ultra 256 GB | MXFP4 | 128K | 84 tok/s | 2.6 s |
| M5 Ultra 512 GB | MXFP4 | 128K | 103 tok/s | 0.8 s |
| H100 SXM | MXFP4 | 128K | 269 tok/s | 0.2 s |
| DGX Station GB300 | MXFP4 | 128K | 346 tok/s | 0.1 s |
| 8× H200 NVL | MXFP4 | 128K | 308 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run gpt-oss:120b- llama.cpp
llama-server -hf unsloth/gpt-oss-120b-GGUF- vLLM
vllm serve openai/gpt-oss-120b- MLX (Mac)
mlx_lm.chat --model mlx-community/gpt-oss-120b-MXFP4-Q4
Questions
How much VRAM does gpt-oss 120B need?
gpt-oss 120B needs 59.0 GB for its weights as released (MXFP4), 60.2 GB with an 8K context and 61.1 GB with 32K.
Can gpt-oss 120B run on an RTX 4090?
Not on one card. The smallest machine here that runs it is the 2× RTX 3090.