OpenAI released GPT-OSS-20B and GPT-OSS-120B, their first open-source large language models under Apache 2.0 license. The 20B model runs on consumer hardware with 16-32GB RAM, while the 120B model requires data center GPUs. Both feature chain-of-thought reasoning, configurable effort levels, fine-tuning support, and 128k token context windows. Users can test them via OpenAI's playground, HuggingFace, or API providers, and run them locally using LM Studio, Ollama, or vLLM. Cloud hosting costs range from $4,000-$12,000 monthly for dedicated GPU instances, with shared options available for lower costs.
Table of contents
Key Specs and FeaturesWhere to try GPT-OSSRunning GPT-OSS LocallySelf-hosting on CloudConclusionSort: