Part 10/13:
In a shift from proprietary restrictions, OpenAI released two openweight models: GPT-OSS120B and GPT-OSS20B. Unlike typical API-based access, these large models are open-source, downloadable, and runnable on local hardware, fostering experimentation and innovation without API lock-in.
GPT-OSS120B: Large-scale reasoning with 117 billion parameters, optimized to run on an 80GB GPU, delivering near its parent’s O4 Mini performance.
GPT-OSS20B: Smaller but still highly capable, compatible with just 16GB of RAM, matching scores from GPT-03 Mini.