available
Weights
Both checkpoints are downloadable.
Inspect this claim’s source ↗OpenAI / FAMILY RECORD
gpt-oss
Two open-weight text reasoning models with adjustable reasoning effort, structured output, and tool-use support.
Checked release: gpt-oss-120b and gpt-oss-20b
Sources checked 2026-09-19
- gpt-oss-120b and gpt-oss-20b release
- · Primary source
The profile describes this checkpoint, not every release carrying the family name. Developer documentation is not independent reproduction.
THE MODEL / THE TERMS
Know what you are picking up.
- Parameters
- 117B / 5.1B active; 21B / 3.6B active
- Architecture
- Transformer mixture-of-experts with alternating dense and locally banded sparse attention.
- Modalities
- text
- Context
- 128K tokens
- Weights license
- Apache-2.0, with a separate usage policy
- Commercial use
- Apache-2.0 permits commercial use; deployers should also review the published usage policy.
- Memory
- OpenAI states that 120b fits in 80 GB and 20b in 16 GB in the released MXFP4 format.
- Known quantizations
- Official MXFP4
Specifications are developer-reported. Primary release documentation ↗ · All sources and terms
NINE DIMENSIONS / NO OPENNESS SCORE
What is actually available?
Each component has its own evidence. “Not established” means this source review did not verify availability; it is not a claim that nothing exists.
available
Architecture
The model card publishes architecture and parameter details.
Inspect this claim’s source ↗available
Inference code
Reference PyTorch and Metal implementations and Harmony renderers are public.
Inspect this claim’s source ↗not established
Training code
The complete training stack was not released.
Inspect this claim’s source ↗partial
Training recipe
Pre- and post-training approaches are described without a fully reproducible recipe.
Inspect this claim’s source ↗partial
Data information
OpenAI says the dataset is mostly English and text-only, focused on STEM, code, and general knowledge.
Inspect this claim’s source ↗not established
Training data
No sufficiently specific public artifact was confirmed in this review.
Inspect this claim’s source ↗partial
Evaluation
A detailed model card and vendor evaluations are published.
Inspect this claim’s source ↗available
Commercial use
Apache-2.0 permits broad commercial use.
Inspect this claim’s source ↗WHY THIS RELEASE MATTERS
What to explore.
- Tool use
- Adjustable reasoning effort
- Local deployment
- Structured outputs
WHAT THE RECORD DOES NOT ESTABLISH
Keep the limits.
- Text-only
- Training corpus is not released
- Self-hosters own system-level safeguards
OpenAI publishes extensive internal evaluations; independent reproductions are not incorporated in this record.
CONNECTED EVIDENCE
A release becomes a starting point.
- Earlier Bright evidence: gpt oss
- Running a reasoning model on your own machine
OpenAI's gpt-oss family provides downloadable reasoning-model weights for self-hosted tool use, structured outputs, and adjustable reasoning effort.
Apache-2.0 weights and public inference references do not include the full training data or recipe. Benchmark and memory figures are vendor results tied to the stated quantization and workload.
Thematic or family connections, not proof that an older application uses the selected checkpoint.
FOLLOW THE RECORD
Sources, not superlatives.
- Introducing gpt-oss ↗ (institution)
- gpt-oss model card ↗ (report)
Source check: 2026-09-19. AI-assisted editorial review, with no named external expert or independent hardware validation claimed. Licenses and model cards can change. Review the linked terms for the specific artifact and intended use.
Understand open source vs open weights ↗