What happened
OpenAI released two new open-weight language models, gpt-oss-120b and gpt-oss-20b, marking its first open-weight release since GPT-2 in 2020. These models are designed for strong reasoning and tool use, and are available under the Apache 2.0 license. However, OpenAI did not disclose key details such as the modelsâ architecture, training data, or methods.
Why it matters
- The release provides developers and researchers access to large-scale open-weight models for the first time in several years.
- These models enable advanced reasoning capabilities while being optimized for efficient deployment on consumer hardware.
- Despite the open-weight release, OpenAI retains proprietary control over critical model components, limiting full transparency.
Key facts
- 1. The gpt-oss-120b model has 116.8 billion parameters and runs efficiently on a single 80 GB GPU.
- 2. The models were released on August 5, 2025, under the Apache 2.0 license.
- 3. The gpt-oss-120b model achieves near-parity with OpenAIâs proprietary o4-mini model on core reasoning benchmarks.
