OpenAI challenges Meta with free customisable AI models

OpenAI has stepped into direct competition with Mark Zuckerberg’s Meta by launching its own set of freely available artificial intelligence models.

The creator of ChatGPT unveiled two open-weight large language models—gpt-oss-120b and gpt-oss-20b-two—which are available for public download and customisation. This marks a notable shift from the company’s flagship ChatGPT product, which operates on a proprietary, closed model.

Meta has already made waves with its Llama series of models, which are similarly open for public use and modification. OpenAI’s latest move signals a growing trend toward broader accessibility in AI development.

"For wide benefit"

Sam Altman, CEO of OpenAI, expressed enthusiasm about contributing to the growing ecosystem of publicly available AI systems, stating the company aimed to promote models “based on democratic values … and for wide benefit.” He acknowledged the investment behind the models, noting they represent “the result of billions of dollars of research,” and said the goal was to place AI “into the hands of the most people possible.”

However, Altman also acknowledged the complexity of making such technology widely available, echoing concerns raised by Meta about the potential risks of releasing highly advanced AI systems to the public.

Mark Zuckerberg has previously justified Meta’s open approach by arguing that widespread access to AI technology helps prevent control from being concentrated among a few powerful tech firms. Other competitors, such as Chinese company DeepSeek, have also introduced powerful, freely customisable models in recent months.

According to OpenAI, the new models outperform other language models of similar size on reasoning tasks. The larger 120b model reportedly achieves results close to the performance of the company’s o4-mini model in core reasoning capabilities.

OpenAI also disclosed that during internal testing, it deliberately created “maliciously fine-tuned” versions of the models to assess their potential misuse in areas like biotechnology and cybersecurity. However, the company found these adapted models were “unable to reach high capability levels.”

Red flags

Despite these assurances, the release of powerful, freely available AI systems has raised red flags among some experts, who warn that such models could be exploited to cause harm, including the potential development of bioweapons.

Meta classifies its Llama models as “open source,” providing access not only to the models but also to their training data, architecture, and source code. However, the US-based Open Source Initiative has disputed this label, arguing that Meta’s usage restrictions disqualify it from being fully open source.

In contrast, OpenAI uses the term “open weight,” a designation that allows model fine-tuning but stops short of complete transparency, as the full workings of the models are not revealed to developers.

GPT-5 about to be released

OpenAI’s announcement comes amid growing speculation about the imminent release of GPT-5, the next generation of its core AI model. Over the weekend, Altman shared a screenshot suggesting a new model may soon be unveiled.

Meanwhile, Google is advancing its own AI ambitions. The company recently revealed Genie 3, a “world model” designed to simulate realistic environments for training AI agents such as robots and autonomous vehicles. Developed by Google DeepMind, the model represents a step toward artificial general intelligence (AGI)—AI capable of performing a broad range of human-like tasks rather than just narrow ones like translation or game playing.

“We expect this technology to play a critical role as we push toward AGI,” said DeepMind, highlighting the increasing importance of AI agents capable of interacting autonomously with the world.

Follow me on X in order to fight your AI FOMO.