gpt-oss 20B (MXFP4)
OpenAI's open-weight mixture-of-experts model for chat and reasoning.
Sourced directly from the model's publisher on huggingface.co. Please double-check links and verify before downloading.
- source https://huggingface.co/mlx-community/gpt-oss-20b-MXFP4-Q8
- license apache-2.0
- base gpt-oss 20B
- published 2026-10-09 by MLX Community staff
> about this model
gpt-oss 20B from OpenAI, converted to MLX by mlx-community in MXFP4 with 8-bit attention and router layers. An open-weight mixture-of-experts model with 21B parameters in total and 3.6B active per token, for chat and reasoning. Text only.
> before you run it
Where did these weights come from?
The source link at the top of this page. Every catalog entry is published by a site admin and must carry a resolvable origin. There is no user upload path anywhere on this site.
Can I verify what I downloaded?
No checksum is published for this entry yet. Verify against the source repo's own file hashes before running it.
Will it fit in my memory?
Budget roughly 17 GB of unified memory. Context length and batch size push that up.
Is this hosted here?
No. We catalog and link; the bytes come from the origin repo. That keeps the chain of custody with whoever actually published the weights.