Publisher memory claim
1 × 16 GB-class device
Native mixed MXFP4
Compatible MXFP4 runtime + Harmony
Publisher states 16 GB of device memory. This is not a certification of every 16 GB GPU or its available kernels.
gpt-oss / MoE
gpt-oss-20b
A compact reasoning baseline that still earns its place
New to model sizes or GPU memory? Start with weights, parameters and quantization →
The older release remains useful when local memory and a controllable reasoning budget matter more than recency. Its publisher-documented 16 GB memory class makes it a practical baseline before moving to larger 2026 systems.
21Breported total parameters
3.6Bactive parameters per token
131,072configured token positions
Publisher-reported, rounded model count; not an exact tensor census. Multimodal components and packaging can change the artifact size.
Evaluate it for
Local text reasoning, tool-use prototypes and experiments with reasoning effort.
Choose another path when
Image input or a runtime that treats Harmony as an ordinary chat template and exposes internal channels incorrectly.
Release context
Why this release matters
The August 2025 release brought open-weight reasoning, Harmony formatting and mixed MXFP4 expert storage together. Those integration choices remain relevant even when you select a newer model.
Deployment starting points
Which hardware, how many, at what precision?
Start with an exact artifact and an engine that supports it. A configuration below is evidence of a documented path; its device count does not promise a particular throughput or concurrency.
13.76 GB of tensor data in the pinned index
This is the index’s reported tensor payload, in decimal GB. It excludes file headers and runtime memory. Mixed precision, conversion and host/device placement determine how much GPU memory the loaded model needs.
Inspect the exact index metadata →Capacity screen: would the weights fit?
This estimates inference memory for a hypothetical uniform precision. It is useful for rejecting an allocation that is too small. It does not establish a working deployment or estimate training memory.
A uniform-precision GPU count would hide this model’s component layout. Publisher-reported, rounded model count; not an exact tensor census. Multimodal components and packaging can change the artifact size. Use the documented artifact and placement path above; no GPU count is inferred here.
Architecture in practice
What the serving engine has to do
Local and global layers read different spans.
Window length and layer ratio are checkpoint-specific.
Full-context state remains for global layers.
Mechanism schematic based on the pinned model card and configuration. It explains a design principle; it is not a full implementation graph.
Inspect the full checkpoint specifications
| Exact checkpoint | openai/gpt-oss-20b |
|---|---|
| Text model type | gpt_oss |
| Layers | 24 |
| Hidden width | 2,880 |
| Attention / KV heads | 64 / 8 |
| Head dimension | 64 |
| Routed / selected experts | 32 / 4 |
| Configured positions | 131,072 |
| Documented extension | Not recorded for this checkpoint |
| Inputs → output | text → text |
| License metadata | apache-2.0 |
12 sliding attention layers; 12 full attention layers. These counts describe the configured pattern.
Open weights do not imply unrestricted use. Read the applicable license terms linked from the pinned card.
A useful comparison
What else belongs on the shortlist?
120b isolates scale within the same release; GLM Flash provides a newer compact MoE alternative.
5 Aug 2025
gpt-oss-120b
A single large-memory GPU is a useful deployment class
117B reported total · text
19 Jan 2026
GLM-4.7-Flash
A compact way to learn latent-attention serving
30B reported total · text
These are editorial comparison candidates. We have not run a matched quality or serving benchmark, so this is not a ranking.
Prepare to explain it
The interview lesson
Explain how a larger reasoning budget changes output length and latency, and why MXFP4 experts do not mean every tensor uses four bits.
Attention layouts · Quantization · Courses and worked examples
For a deployment evaluation, record exact weights, precision, engine, device count, interconnect, prompt/output lengths and concurrency. Report task success, errors, TTFT, TPOT and useful throughput together.
Sources and citation
Our assessment is editorial judgment based on the linked architecture and deployment evidence. Model facts are publisher-reported or attributed to runtime maintainers; no independent GPU benchmark was run.
AI Infra Interviews, “gpt-oss-20b: release, architecture and deployment”, checked 2026-09-13. Preserve this date and the exact checkpoint when citing.