OpenAI presents new powerful AI models: o1 and o1-mini

The AI startup OpenAI, supported by Microsoft, has unveiled a new generation of artificial intelligence models capable of handling complex tasks similar to humans. The new models, o1 and o1-mini, have been made available recently and are accessible via ChatGPT as well as through their API. According to a company blog post, the models were trained to spend more time on problem analysis, akin to human thinking. Through this training process, the models learn to refine their thought process, try various strategies, and recognize errors. An OpenAI co-founder has raised $1 billion for a startup dedicated to developing a safe superintelligence. The new models are capable of solving more complex tasks in science, programming, and mathematics compared to their predecessors. The blog post reported that the o1 model achieved 83% of the qualification points for the International Mathematical Olympiad, compared to 13% by the previous model GPT-4o. The models’ capabilities are also evident in competitive programming questionnaires. Additionally, they surpassed PhD-level accuracy in scientific benchmark tests. OpenAI CEO Sam Altman praised the new model series on X, describing them as "the most capable and synchronized models to date." He added, "o1 is still flawed, still limited, and seems more impressive on first use than over time. However, it also marks the beginning of a new paradigm: AI that can perform general complex problem-solving.