Can I run gpt-oss 20B locally?
Yes. At MXFP4 with 8K context, gpt-oss 20B needs about 14 GB, so it runs entirely on any GPU with 15 GB of VRAM or more, starting with the AMD Radeon RX 9060 XT 16GB.
Memory needed by compression and context length
| Compression | 4K context | 8K context | 32K context | 128K context |
|---|---|---|---|---|
MXFP4 The format the model is released in. | 14 GB | 14 GB | 15 GB | 17 GB |
Total memory: model weights, context (KV cache) and about 1 GB of runtime overhead. To run entirely on the GPU, VRAM must be about 5% larger than this.
GPUs that run gpt-oss 20B entirely on the GPU
- Amazon
AMD Radeon RX 9060 XT 16GB
16 GB · ~83 tok/s at MXFP4
- Amazon
AMD Radeon RX 7600 XT
16 GB · ~76 tok/s at MXFP4
- Amazon
AMD Radeon RX 9070 XT
16 GB · ~147 tok/s at MXFP4
- Amazon
AMD Radeon RX 9070
16 GB · ~147 tok/s at MXFP4
- Amazon
AMD Radeon RX 7800 XT
16 GB · ~145 tok/s at MXFP4
- Amazon
AMD Radeon RX 9070 GRE
16 GB · ~107 tok/s at MXFP4
- Amazon
NVIDIA GeForce RTX 5060 Ti 16GB
16 GB · ~111 tok/s at MXFP4
- Amazon
NVIDIA GeForce RTX 4060 Ti 16GB
16 GB · ~76 tok/s at MXFP4
- Amazon
AMD Radeon RX 6800
16 GB · ~123 tok/s at MXFP4
- Amazon
NVIDIA GeForce RTX 4080
16 GB · ~161 tok/s at MXFP4
Model facts
- Parameters
- 20.9B (3.6B active, mixture-of-experts)
- Max context
- 128K tokens
- License
- Apache 2.0
- Released
- Aug 2025
Check it with your hardware
Change the compression, context length and system RAM in the calculator.
Open calculator