The cost to use conversational AI models is plunging, and developers are finding new uses for the technology, from AI coding assistants to “model routing” startups that help customers choose the cheapest model for each use case. The cost to use a GPT-4–level AI model from OpenAI has dropped more than 85% since early 2023. The decline reflects efforts by AI startups to make their models more efficient and therefore cheaper to run. As the companies draw more customers, they also use the hardware they’re renting more efficiently, via a strategy called “batching.” But the trend may not continue in 2025. In December, OpenAI announced a pro version of its chatbot that costs a whopping $200 per month—10 times as much as the current premium version of ChatGPT. The offering is aimed at users who want access to a version of OpenAI’s reasoning model that can answer more difficult questions by taking more time to calculate an answer. Weeks later, the company revealed the next generation of its reasoning AI, o3, which posted impressive results; but researchers said it cost them more than $1,000 to complete each task. With the rise of reasoning models, 2025 could become the year of the haves and have nots in AI, where well-heeled developers can afford to let reasoning models work on tough problems, a process that could take days or weeks, while others won’t have that opportunity. #Data is beautiful