This week, OpenAI announced gpt-oss, a new AI model designed for local use, alongside the much-anticipated GPT-5. Unlike GPT-5, gpt-oss is “open-weight,” allowing developers to customize it for specific applications. However, it isn’t fully open source because it lacks accessible underlying code and training data. For general users, gpt-oss offers the ability to run a large language model (LLM) offline, enhancing privacy by keeping data local.
There are two versions available: gpt-oss-20b and gpt-oss-120b, with the latter requiring substantial hardware. Installation on a Mac is straightforward via the Ollama app. Tests revealed that gpt-oss significantly lagged behind ChatGPT in response time, yet it excels in user privacy, as it processes queries without sending data to OpenAI. This makes gpt-oss a potentially appealing option for users with powerful computers seeking a private AI solution.
Source link
