OpenAI · open-weights model
gpt-oss-20b
gpt-oss-20b is a 21B-parameter model (MoE, 3.6B active) by OpenAI. Its smallest quant (MXFP4) needs at least 13.0GB of GPU-addressable memory at 8K context; it loads cleanly at a quality quant on 36 of the 50 machines we track.
- Parameters
- 21B total · 3.6B active
- Layers
- 24
- Max context
- 128K
- Attention
- HYBRID
- License
- apache-2.0
Config from openai/gpt-oss-20b
Quants and what they need
Measured file sizes; “needs” = weights + KV cache at 8K context + overhead.
| Quant | File size | Needs ≥ |
|---|---|---|
| MXFP4 | 12.1 GB | 13.0 GB |
What runs gpt-oss-20b
Every machine we track, graded at its best clean-fit quant.
Verdicts computed from published specs and measured GGUF sizes — how the math works.