Technology

GPT-OSS

OpenAI's open-weight LLM family (gpt-oss-120b and gpt-oss-20b) delivers state-of-the-art reasoning and agentic capabilities, optimized for efficient local and on-device deployment.

GPT-OSS is OpenAI's open-weight language model family, released under the permissive Apache 2.0 license: a significant shift from their traditional proprietary models. The family features two models, the 117B-parameter `gpt-oss-120b` and the 21B-parameter `gpt-oss-20b`. Both utilize a Mixture-of-Experts (MoE) architecture, enabling high performance with low resource use: the smaller model runs on consumer hardware (16GB RAM) while the larger fits on a single 80GB GPU (e.g., NVIDIA H100). These models excel at complex reasoning and agentic tasks (function calling, web browsing), providing developers with full chain-of-thought access for flexible, high-performance local inference and custom application deployment.