GPT-5.2 series consistently score higher than Opus 4.5 on Art AGI (arcprize.org)

The latest evaluation of AI models on the ARC-AGI-2 platform reveals that the GPT-5.2 series consistently outperforms Opus 4.5, highlighting a significant advancement in the landscape of artificial general intelligence (AGI). ARC-AGI has transitioned from assessing basic fluid intelligence in its initial version to evaluating adaptability and efficiency in complex tasks. This comparison is critical as it emphasizes that true intelligence encompasses not only problem-solving capabilities but also the efficiency of resource utilization, measured by cost-per-task. These results carry substantial implications for the AI/ML community, particularly as they suggest that the GPT-5.2 series may lead to more cost-effective and capable systems in practical applications. The ability to achieve high performance while maintaining lower operational costs positions these models favorably in an increasingly competitive field. As the testing is ongoing and provisional cost estimates based on models like Gemini 3 Pro are in flux, the community eagerly anticipates more comprehensive results, which could redefine benchmarks for efficiency and performance in AI systems.