gpt-oss-20b
gpt-oss-20b is OpenAI's open-weight reasoning model built for local, on-device, and specialized use, with adjustable reasoning and harmony format.
Model Overview
Capabilities, design details, and architectural traits
gpt-oss-20b - OpenAI's Open-Weight Reasoning Model for Local and On-Device Use
gpt-oss-20b is an open-weight reasoning model released under the Apache 2.0 license. OpenAI positions it as the smaller of two gpt-oss models, built for lower latency, local, and specialized use rather than large-scale production serving.
| Trait | Detail |
|---|---|
| Defining purpose | Positioned by OpenAI for lower latency, local, and specialized use cases, as the smaller of the two gpt-oss models. |
| On-device deployment | Documented as ideal for on-device use, local inference, or rapid iteration without costly infrastructure, rather than large-scale production serving. |
| Format requirement | Trained on OpenAI's harmony response format; OpenAI states it should only be used in that format, since it will not work correctly otherwise. |
| Agentic tool use | Supports adjustable reasoning effort across low, medium, and high settings, and is built for agentic workflows with tool use such as web search and Python code execution. |
A Different Risk Profile Because It Is Open Weight
OpenAI states that open-weight models like gpt-oss-20b carry a different risk profile than its proprietary models. Once released, a determined attacker could fine-tune the weights to bypass safety refusals, and OpenAI would have no way to add further mitigations or revoke access after the fact.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | OpenAI | Anthropic | Anthropic |
| Release Date | August 5, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | Jun 2024 | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 131K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.07 Best Input Pricing | $4 | $2 |
| Output Pricing | $0.18 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 9.0 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 20.7 | - | - |
| Agentic Index | 1.2 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from OpenAI
Other models by OpenAI
Top AI Models
Leading alternatives by intelligence score
