OpenAI vs. Meta: Free AI Models Are Changing the Game in Artificial Intelligence

OpenAI, the company known for developing ChatGPT, has announced the launch of two freely available artificial intelligence models that can be customized to developers' needs. These models, called gpt-oss-120b and gpt-oss-20b, represent a significant departure from OpenAI's previous approach, in which ChatGPT was based on a closed model that could not be modified. This move puts OpenAI in direct competition with Mark Zuckerberg's Meta and Chinese rival DeepSeek, which have previously offered similar freely available models. It is an important shift intended to expand access to advanced artificial intelligence to the widest possible audience.

Sam Altman, CEO of OpenAI, expressed enthusiasm about the move and emphasized that the models are the result of billions of dollars invested in research. According to him, these tools are intended to serve the "broader benefit" and be based on democratic values so that artificial intelligence reaches as many people as possible. Like Meta's Mark Zuckerberg, he argues that freely available and customizable models will ensure that the benefits of artificial intelligence are not concentrated in the hands of just a few companies but are accessible to everyone around the world.

Details of the New Models

The gpt-oss-120b and gpt-oss-20b models are described as "open weight," meaning that their trained parameters are publicly available. This allows developers to download, modify, and run them independently. The larger gpt-oss-120b model is designed to run on a single Nvidia graphics card, while the smaller gpt-oss-20b is also suitable for standard laptops with 16 GB of memory. Both models outperform similarly sized competitors on tasks requiring reasoning, with gpt-oss-120b achieving performance nearly equal to OpenAI's internal o4-mini model in basic reasoning.

These models are available for download on platforms such as Hugging Face and are integrated into AWS (Amazon Web Services), enabling businesses and individuals to deploy them with advanced security, customization, and scaling features. OpenAI emphasizes that the models are designed for use in agentic workflows (workflows involving autonomous agents), where they can operate independently, for example as the foundation for AI agents that perform tasks without human intervention. During testing, OpenAI created "malicious" versions of these models simulating biological and cyber threats but found that they did not reach a high level of capability.

Battle with the Competition

OpenAI's move comes at a time when the company is facing pressure from both the industry and the public for greater transparency and customizability. Since the release of GPT-2 in 2019, OpenAI has focused on proprietary models, making this its first major open release in more than five years. This puts OpenAI in direct competition with Meta, which has long promoted an open approach through its Llama models, and with rapidly growing Chinese competitors such as DeepSeek. Meta describes its models as "open source," which includes providing training data, architecture, and code, but the Open Source Initiative points out that Meta's restrictions prevent them from fully qualifying as open source.

Mark Zuckerberg warned that a similar approach would need to be handled carefully with highly advanced models to prevent misuse. OpenAI is taking the more moderate "open weight" approach, under which the models can be modified, but their internal workings are not fully transparent. This shift is seen as a response to growing pressure to democratize artificial intelligence, with models such as those from DeepSeek having already been made available for free download and customization.

Risk Concerns

Although these models promise broader access, they are raising concerns among experts. According to The Guardian and other sources, there is a risk that freely available and customizable models could be modified to cause serious harm, such as developing biological weapons or other threats. OpenAI conducted tests showing that such malicious modifications did not reach a high level of capability, but experts warn of potential risks. Meta similarly acknowledges the need for caution with advanced systems.