gpt-oss-20b
Language modelOpen weightsCard summarizedMedium-sized open-weight model for low latency
● Re-checked 12 Oct 2026 · 102 facts · 0 correctionsSource: gpt-oss-120b & gpt-oss-20b Model Card (35 pp)
Use it via API
Where to call it
- Self-host (weights)
openai/gpt-oss-20b - OpenAI API docs
gpt-oss-20b
Supports
- Streaming
- Structured outputs
- Reasoning control
- Tool calling
- Web search
- Fine-tuning
Good for
- Lower-latency, local or specialized use cases
- Web browsing (search/open), Python in a stateful Jupyter environment, and developer-defined functions
- Adjustable reasoning effort for tasks that don't need deep reasoning
Overview
gpt-oss-20b
Medium-sized open-weight model for low latency
- Cost
- $0.03 host
- Context
- 128K
- Cheapest host
- $0.03 / $0.14
- Max output
- 128K
- Knows up to
- June 2024
- Released
- 5 Aug 2025
gpt-oss-20b is the smaller of OpenAI's two open-weight reasoning models: a text-only Mixture-of-Experts transformer with 20.9B total and 3.6B active parameters, released under Apache 2.0, that runs in as little as 16 GB of memory. It has the same agentic features as gpt-oss-120b (tool use, Structured Outputs, adjustable reasoning effort, full chain-of-thought). OpenAI calls it competitive despite being about 6x smaller; it trails the 120b on knowledge-heavy tasks such as GPQA. §1, §2, §2.6.1
openai/gpt-oss-20b- Input / output per 1M
- $0.03 / $0.14 via DeepInfra
- Context window
- 128K
- Max output
- 128K
- Knows up to
- June 2024
- Reasoning
- Configurable (low / medium / high)
- Weights
- Open · Apache 2.0 (plus gpt-oss usage policy)
About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses OpenAI's gpt-oss-120b & gpt-oss-20b Model Card. Every claim carries its source page.