{"id":43452,"title":"OpenAI rivals Meta and DeepSeek as it launches free customisable AI models","publisher":"Stockmark.IT","author":"Stockmark.IT Website","published":"2025-08-06T05:51:08+00:00","modified":"2025-08-06T05:51:08+00:00","canonical_url":"https://stockmark.it/openai-rivals-meta-and-deepseek-as-it-launches-free-customisable-ai-models/","markdown_url":"https://stockmark.it/openai-rivals-meta-and-deepseek-as-it-launches-free-customisable-ai-models.md","json_url":"https://stockmark.it/openai-rivals-meta-and-deepseek-as-it-launches-free-customisable-ai-models.json","category":"Artificial intelligence","categories":["Artificial intelligence","Meta"],"featured_image":"https://i0.wp.com/stockmark.it/wp-content/uploads/openai.jpg?fit=1200%2C800&quality=89&ssl=1","format":"news","language":"en-GB","content":"OpenAI, the developer behind ChatGPT, has embarked on a new competitive strategy by introducing freely available artificial intelligence models for developers to download and customise. This move positions OpenAI alongside Meta and Chinese competitor DeepSeek, both of which have already released powerful AI systems accessible for public adaptation.\n\nPreviously, OpenAI’s flagship product ChatGPT operated under a closed model, restricting customisation beyond its original parameters. Now, with the unveiling of two “open weight” large language models, dubbed gpt oss 120b and gpt oss 20b two, the firm is signalling a shift toward greater accessibility. According to Sam Altman, chief executive, this initiative aims to “get AI into the hands of the most people possible”, reflecting a wider trend towards democratising the technology.\n\nMeta has also promoted a vision of widely available AI, encouraging broad adoption by making its Llama models downloadable and customisable. However, industry bodies have noted Meta’s licensing framework stops short of proper open source, while OpenAI’s “open weight” terminology indicates the models can be fine-tuned, albeit without full transparency of underlying workings.\n\nOpenAI claims its largest open weight model matches the performance of its o4 mini model in reasoning tasks, outstripping other models of similar scale. Testing included attempts to fine-tune the systems for malicious use, in which researchers reported the models were unable to achieve advanced harmful behaviours — a nod to ongoing concerns amongst experts about the risks of unrestricted AI.\n\nThe release comes at a time when Google’s DeepMind is advancing efforts towards artificial general intelligence with a new world model capable of simulating real world environments. Google hopes such models will one day train autonomous robots and agents, raising the stakes for all players in the industry.\n\nExpressing ambitions rooted in democratic values and wide benefit, OpenAI and its rivals are setting the agenda for the next phase of artificial intelligence innovation. As the global market adapts to increasing transparency and developer empowerment, the contest between OpenAI, Meta and DeepSeek could define industry standards for years to come."}