OpenAI has announced two free, customisable artificial intelligence models, directly challenging Meta and Chinese rival DeepSeek. The ChatGPT developer released 'open weight' large language models, which can be downloaded and fine-tuned by developers, marking a shift from its proprietary ChatGPT system.
Sam Altman, OpenAI's chief executive, said the company was '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'. The models, named gpt-oss-120b and gpt-oss-20b-two, outperformed similarly sized models on reasoning tasks, with the larger version achieving near-equal performance to OpenAI's o4-mini model.
Meta has long offered its Llama models on a similar basis, with Mark Zuckerberg arguing that freely available AI ensures wider access and prevents power concentration. However, Meta has cautioned about applying the same approach to highly advanced AI. DeepSeek, a Chinese competitor, also provides powerful customisable models.
OpenAI stated the models could underpin autonomous AI agents and were 'designed to be used within agentic workflows'. During testing, the company created 'maliciously fine-tuned' versions to simulate biological and cybersecurity threats but found they were 'unable to reach high capability levels'. Experts have warned that freely available AI could be adapted for harmful purposes, such as developing bioweapons.
The term 'open weight' denotes a model that can be fine-tuned but lacks full transparency, unlike 'open source' which also provides training data and code. The Open Source Initiative has said Meta's restrictions mean its Llama models do not qualify as fully open source. OpenAI's announcement coincided with speculation about a new ChatGPT model, with Altman sharing a screenshot of what appeared to be GPT-5.
