Post

Conversation

Updated this chart with the newest Gemini. It shows the rapid progress in AI over less than two years: costs for GPT-4 class models has dropped 99.7% and even the most advanced models in the world are still 82% cheaper. Probably not worth betting on this trend ending really soon
The image is a chart titled "Shifting Frontier of AI Model Performance and Cost" which illustrates the rapid progress in AI over less than two years. It shows the relationship between the cost per million tokens (on a log scale) and the GROA Diamond Score, which measures capability. Key points on the chart include various AI models like Gemini 2.5, Gemini 1.5 Flash, and different versions of Claude and GPT-4, highlighting significant cost reductions and performance improvements. For instance, the cost for GPT-4 class models has dropped by 99.7%, and even the most advanced models are 82% cheaper. The chart uses different colored lines to represent the cost-performance frontier over time, with annotations indicating specific models and their release dates. The post text by Ethan Mollick emphasizes the rapid progress and cost reduction in AI, suggesting that this trend is likely to continue.