each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.
each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.
On open weightsToday, we’re releasing a research preview of gpt-oss-safeguard, our open-weight reasoning models for safety classification tasks, available in two sizes: gpt-oss-safeguard-120b and gpt-oss-safeguard-20b. These models are fine-tuned versions of our gpt-oss open models and available under the same permissive Apache 2.0 license, allowing anyone to use, modify, and deploy them freely. Both models can be downloaded today from Hugging Face.
The post describes the two models as fine-tuned versions of the gpt-oss models and says they are under the same Apache 2.0 license. The license file quoted here is the LICENSE in openai/gpt-oss, opened with the August 5, 2025 release.
each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.
On open weightsWe’re releasing gpt-oss-120b and gpt-oss-20b—two state-of-the-art open-weight language models that deliver strong real-world performance at low cost. Available under the flexible Apache 2.0 license
The post says “gpt-oss models are our first open-weight language models since GPT-2” and that the weights are “freely available for download on Hugging Face.” On the same page: “Once an open-weight model is released, adversaries may be able to fine-tune the model for malicious purposes.” It says that even with robust fine-tuning, maliciously fine-tuned models “were unable to reach high capability levels” under the Preparedness Framework, and that “These findings informed our decision to release the gpt-oss models.”
each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.
Maple runs open-weight models inside a hardware-isolated enclave. Maple cannot see your chats. How it works
Every entry cites one primary source, quotes the document’s own words about open weights, and is checked against that source before it publishes. Corrections keep the old wording visible. Methodology · Corrections · Suggest an entry or a correction
