The Age of Abundant Intelligence

Throughout the history of generative AI, the default assumption has usually been that if you want the smartest AI, you need to pay for the biggest and most expensive models from the largest labs. Simply put, the best models were proprietary, expensive, and meaningfully more intelligent than the alternatives. Most people, even today, still carry that mindset. What they don’t realize is that relationship is starting to change. Every month, new models become available that are faster and cheaper, and are closing the gap on the largest labs’ best offerings.

This becomes instantly apparent when you look at models being released outside of the traditional frontier labs. Open-weight models from companies like DeepSeek, Zhipu, and Alibaba are becoming increasingly capable of handling work that, until recently, would have required expensive proprietary models to accomplish. This trend isn’t limited to just open models, either. Newer models from SpaceXAI and Meta are also competing aggressively on the price-to-performance front. Now, this doesn’t mean that all models are created equal, or that the frontier no longer matters. It just means that for an increasingly large percentage of real-world tasks, you can still get very strong results without the most expensive model money can buy.

This fundamentally shifts how you should think about using AI. If your typical workflow involves document summarization, topical research, data analysis, software authoring, classification, or automation, you probably don’t need the world’s smartest model for every step. It’s becoming increasingly clear that the better, more effective approach is to use inexpensive models for the bulk of your tasks, and reserve premium models for the smaller number of problems where their additional capabilities actually matter. As the cost of useful AI approaches pennies, or even fractions of pennies, the question is no longer whether something is important enough to justify the use of AI, but rather why not?

Services like OpenRouter let you choose from hundreds of models and providers, trading off intelligence, speed, and cost. Developers can get access to capable models through services like OpenCode Go, while Hugging Face provides an enormous ecosystem for exploring and running open-weight models. Just a few short years ago, access to state-of-the-art intelligence was exclusive to the offerings of private labs. Today, we’re increasingly able to pick and choose intelligence like any other computing resource. For me, that’s what makes the age we’re living in so exciting. We already have access to models that are becoming more capable, more affordable, and more accessible by the day. People are using them right now to build incredible products, automate complex workflows, and do things that would have required far more time, money, or expertise just a few years ago. And as the barriers keep falling, more people will have the opportunity to build with this level of intelligence themselves.