OpenAI researchers have admitted that even the most advanced AI models can't really solve the coding problems put in front of them. In a new paper that's awaiting peer review, the company's researchers acknowledged that even frontier models, or the most advanced and boundary-pushing AI systems, "are still unable to solve the majority of [coding] tasks." Using their newly-developed SWE-Lancer benchmark, which was built on more than 1,400 software engineering tasks from the Upwork freelancer network (hence the moniker). To evaluate how well they performed, OpenAI put three large language models (LLMs) — OpenAI's o1 reasoning model and its GPT-4o, […]