Open-weight models are becoming a bigger part of the AI market, and they can already be the better choice for some of the most advanced tasks. For a particular job, the frontier may now be an open-weight model.
Anthropic’s slowing growth does not mean the market is not expanding. The pattern in Ava’s briefing tells us that we need to look beyond any single provider to understand where demand is going.
Ava highlights two shifts in the economics of AI. On Vercel’s AI Gateway, open-weight usage rose from 10% to 81% in five months: a gain of 71 percentage points. The token-price data shown in the video records a 53% fall over four months.
At Kydosa, we feel the edges of AI on real work every day. As Ava points out, safety training can make some models overthink. The best choice depends on the work. We need to keep testing which models produce the most useful results, including on our most demanding tasks.
We see OpenAI’s stronger tools around its models and rising capacity pressure pointing to its next growth leg. The models, the tools and the economics are all moving, and our choices will move with them.
Our bigger bet is Jevons’s paradox. As AI gets cheaper, demand for the work we do goes up. Cheaper, more capable AI will drive more demand. Businesses will find more and more ways to use it, and turn it into useful results.
Our expectation is that the overall demand for knowledge work (human + machine) will increase exponentially. Helping businesses turn that growing capability into useful results is the opportunity we see ahead.