In 2019, researcher François Chollet proposed an innovative benchmark for artificial intelligence: ARC-AGI. At that time, the idea seemed ahead of its time, as there were no models capable of tackling such tests. Chollet anticipated the rise of AI, which would come years later with the emergence of ChatGPT and the subsequent boom in the sector. Over time, numerous synthetic benchmarks emerged to evaluate the performance of AI models, but ARC-AGI stood out for its approach: instead of measuring memorization ability, it tested abstract thinking and generalization. The challenges of ARC-AGI and its later version, ARC-AGI 2, consist of visual puzzles that are simple for humans but have traditionally been a hurdle for machines. However, in the last two years, AI models have made significant advancements in these areas, increasingly solving ARC-AGI problems. In 2024, the model o3-preview managed to solve 87% of ARC-AGI 1, a feat that the benchmark's creators publicly highlighted. However, the cost was high: $456,000 for 100 tasks, or $4,560 per task. Yesterday, OpenAI introduced GPT-5.2, its latest model. While its performance on other benchmarks was already outstanding, the most striking aspect was its performance on ARC-AGI 1: GPT-5.2 Pro (X-High) solved 90.5% of the problems, but what was truly revolutionary was the cost, which dropped to just $11.65 per task, 390 times less than the previous year. Even a more economical version, GPT-5.2 (X-High), achieved an 86.2% success rate for only $0.96 per task. Anticipating the progress of AI, Chollet and his team launched ARC-AGI 2 in March 2025, an even more demanding version. So far, the best result had been achieved by Claude Opus 4.5, solving only 38% of the problems. GPT-5.2 made a significant leap, reaching nearly 55% accuracy while maintaining a cost of $15.72 per task. The trend is clear: AI is not only improving in capability but also becoming much more economically accessible. This is especially relevant in a context where scaling models no longer produces spectacular jumps in performance, but rather in efficiency and cost. The AI industry seems to be at a turning point. The question is no longer whether AI can solve a problem, but how much it will cost to do so. GPT-5.2 demonstrates that it is possible to advance on both fronts: greater capability and lower cost, a crucial factor for OpenAI in a time of economic challenges. Moreover, this advancement represents a direct response to competition, such as Gemini 3 Pro, and paves the way for a more efficient and affordable AI.