Over the course of 19 months, the cost to reach 75% on ARC-AGI-1 fell 99.95% (twitter.com)

In a significant development for the AI/ML community, the cost to reach 75% efficiency on the ARC-AGI-1 model decreased an astounding 99.95% over 19 months, dropping from $26 per task using the o3-preview to just $0.01 per task with the DeepSeek V4 Flash. Similarly, the costs associated with the ARC-AGI-2 model showed a remarkable 99.44% reduction in just six months, falling from $13.62 per task with Gemini 3 Deep Think to $0.08 per task using the Dots3-Note Preview. These dramatic cost reductions highlight significant advancements in AI model optimization and operational efficiency, making powerful AI solutions more accessible. The decreasing expense per task can accelerate the adoption of these models across various industries, facilitating widespread experimentation and innovation. The rapid improvements also raise implications for computing resource allocation, enabling researchers and developers to leverage high-performance AI capabilities at unprecedented affordability.