Artificial Intelligence (AI) has made significant strides in recent years, but the question remains: where are its limits? OpenAI’s latest model, known as 03, showcases notable improvements over its predecessors, although it comes with high financial costs. This new AI utilizes a technique called Test-Time Compute, which allows the model to take more time to generate solutions, aiming to provide better answers to complex questions. However, it is not yet considered a general artificial intelligence (AGI).
In a benchmark test comparing artificial intelligence with human intelligence, the 03 model performed surprisingly well. By the end of December 2024, François Chollet, the developer of the test, announced on social media that the model achieved 87.5 percent in the benchmark. To put this into perspective, this score is nearly three times higher than the previous model, o1, which scored 32 percent.
Despite the improvement, the cost of operating the 03 model is substantial. Chollet noted that running the system at full capacity could cost thousands of dollars per task. Even the less powerful version of 03 incurs a cost of 20 US dollars per task, with a slightly lower accuracy of 75.7 percent.
Within the industry, there has been ongoing debate about whether enhanced AI models will eventually hit a ceiling due to their increasing demand for computational power and training data. The technical performance in tests suggests that these concerns might be unfounded. However, the escalating costs per task raise questions about the viability of such advancements and whether financial constraints will limit their adoption.
For example, the less advanced version of 03, costing 20 dollars per task, still represents a significant expense compared to previous models or the monthly 20 dollars charged for ChatGPT Plus. These costs fuel discussions about the sustainability and accessibility of advanced AI technologies.
Looking ahead, a mini-version of the 03 model is planned for release in January 2025. Currently, the 03 model is not publicly available, but researchers can apply for early access to explore its capabilities.
As AI continues to evolve, the balance between performance and cost will be crucial in determining its future applications and accessibility. The industry must address these challenges to ensure that AI advancements remain beneficial and accessible to a broader audience.