GPT models are once again playing the “cost-performance” card: Luna is down 80%, Terra is down 20%, and Sol has also added faster API options. On the surface, they’re just adjusting prices, but what I care about is another layer—using the same amount of compute in Codex and ChatGPT Work can take you much farther.
My understanding is simple: capabilities are still moving forward, but marginal costs are being driven down quickly. For people building applications, the threshold for trial and iteration is getting lower. For the industry as a whole, the competitive focus will gradually shift from “who has the strongest model” to “who can use the model more cheaply, faster, and more reliably.”
On the crypto side, I’ll still keep an eye on AI compute, cloud, and infrastructure-related names, but I won’t treat a single price-cut announcement as a “must-buy” thesis. As models become more like utilities such as water and electricity, what’s truly valuable may still be the scenarios and the data feedback loop (￣▽￣)
My understanding is simple: capabilities are still moving forward, but marginal costs are being driven down quickly. For people building applications, the threshold for trial and iteration is getting lower. For the industry as a whole, the competitive focus will gradually shift from “who has the strongest model” to “who can use the model more cheaply, faster, and more reliably.”
On the crypto side, I’ll still keep an eye on AI compute, cloud, and infrastructure-related names, but I won’t treat a single price-cut announcement as a “must-buy” thesis. As models become more like utilities such as water and electricity, what’s truly valuable may still be the scenarios and the data feedback loop (￣▽￣)
