GPT-oss-120b

GPT-oss-120b represents OpenAI's first major foray into open-weight models, featuring 117B total parameters implemented using a Mixture of Experts (MoE) architecture. This architectural choice allows the model to maintain high capability while keeping inference costs manageable through sparse activation of expert modules.

The model rivals the performance of OpenAI's proprietary o4-mini model, demonstrating that OpenAI can produce competitive open-weight models without sacrificing too much capability compared to their closed offerings. GPT-oss-120b is fully open-weight with commercial use explicitly allowed, removing traditional barriers to deployment and modification.

This release marks a significant strategic shift for OpenAI, which has historically focused exclusively on proprietary API-based models. By releasing a capable open-weight model, OpenAI addresses criticism about AI centralization and provides researchers and developers with access to a high-quality base model for fine-tuning and research. The model's performance parity with o4-mini makes it particularly valuable for applications where self-hosting or model customization is preferred over API-based deployment.