OpenAI has launched its latest open-weight large language models (LLMs), known as GPT-OSS, including gpt-oss-120b and gpt-oss-20b. These “reasoning models” are designed to excel in complex reasoning tasks, emphasizing OpenAI’s mission of creating AI that benefits humanity. Unlike popular closed models such as OpenAI’s own GPT series and Google Gemini, open-weight models allow users to download and run the entire model on their own hardware, though they lack complete training data and documentation like open-source models. The gpt-oss-120b features 117 billion parameters delivering a 90% accuracy rate on the MMLU benchmark, while gpt-oss-20b has 21 billion parameters and achieves 85.3% accuracy. These models require significant system memory: 80GB for gpt-oss-120b and 16GB for gpt-oss-20b. OpenAI aims to enhance innovation, safety, and transparency in AI development through these advancements, returning to their foundational mission amidst increasing competition.
Source link
Share
Read more