OpenAI launches free, customizable large language models to rival Meta and DeepSeek
OpenAI just dropped two new “open weight” large language models—gpt-oss-120b and gpt-oss-20b-two—that anyone can download and customize. This is a sharp pivot from ChatGPT’s closed, unmodifiable model.
The 120 billion parameter model matches ChatGPT’s o4-mini in core reasoning, outperforming similarly sized models. OpenAI says these models are built for “agentic workflows,” meaning they can power AI agents that act autonomously. 
CEO Sam Altman stated:
“We’re excited to make this model, the result of billions of dollars of research, available to the world to get AI into the hands of the most people possible.”
He also framed the effort as rooted in “democratic values” aimed at broad benefit.
Meta, with its Llama open-source models, pushed a similar agenda. Zuckerberg has said free and customizable models help avoid concentrating AI power. But Meta warns it might need to be cautious with more advanced AI models.
China’s DeepSeek also offers powerful downloadable and customizable models, creating a heated global race.
OpenAI tested “maliciously fine-tuned” variants designed to simulate bio-threats and cyberattacks but found they didn’t reach high capabilities—though experts warn freely available models could be misused.
The difference: OpenAI calls its models “open weight,” meaning developers can fine-tune them, but the full workings aren’t entirely transparent. Meta’s Llama models are “open source” but with usage restrictions that the Open Source Initiative says disqualifies them from being fully open.
Altman teased more AI news Sunday by sharing a screenshot hinting at GPT-5’s possible release.
Meanwhile, Google revealed Genie 3, a “world model” for realistic AI simulations aimed at training robots and autonomous systems—another step in its push toward artificial general intelligence (AGI).
DeepMind said:
“We expect this technology to play a critical role as we push toward AGI, and agents play a greater role in the world.”
