The AI industry is on the cusp of a significant shift. For years, tech companies have assumed that bigger models are more powerful and therefore better. However, with mounting costs and rising token prices, users are being forced to reconsider this approach.
According to Coinbase co-founder Brian Armstrong, the answer is a resounding yes. In a recent post on X, he predicted that 80% of workloads will be running on 99% cheaper models within 12-18 months. This would mean a massive shift in the economics of AI, with much of the savings coming out of the pockets of big labs like OpenAI and Anthropic.
But what does this mean for tech companies? Can they really switch to smaller models without compromising quality? Initial tests suggest that it’s possible. A recent test by the legal AI tool Harvey showed that reducing inference costs by 3x did not affect quality. The company combined Claude Opus with Fireworks’ GLM 5.1 and shifted to Opus for the most intensive tasks, resulting in a significantly lower load on servers.
The trend is often framed as a battle between major labs and Chinese models or open-weight ones. However, this misses the bigger point: the real divide is between large models and small ones. There’s an active price war going on between in-house inference from big labs and independently served open-weight models. The key takeaway is that users can save money by switching to smaller models without sacrificing quality.
The industry has been dominated by a scaling-first approach, where labs lean hard into training the most compute-intensive models possible. However, with prices heavily subsidized by investors, clients had no reason to choose anything but the most advanced option. With token prices rising and subsidies slowing down, users are facing cost pressure for the first time.
It’s unclear whether this new cost pressure will drive enterprise users to smaller models or if they’ll find other ways to economize. However, if it turns out that most deployments can be run just as well on a smaller model, it could put a serious damper on the growing demand for inference and raise questions about how to justify the cost of training frontier models.
**The Shift in AI Economics**
For years, tech companies have been driven by a scaling-first approach. They’ve focused on building ever-larger models that can handle increasingly complex tasks. However, this approach has come at a significant cost. The rise of token prices and the slowing down of subsidies are forcing users to rethink their approach.
**The Rise of Smaller Models**
Smaller models have been gaining traction in recent months. They offer a more affordable alternative to large-scale models, without sacrificing quality. Initial tests suggest that they can perform just as well as larger models, but at a fraction of the cost.
**The Price War Between Labs and Open-Weight Models**
The industry is seeing an active price war between in-house inference from big labs and independently served open-weight models. This is driving down costs for users and forcing labs to rethink their approach.
**Conclusion**
The AI cost revolution is here, and it’s time for tech companies to take notice. With smaller models gaining traction and the rise of a price war between labs and open-weight models, users have more options than ever before. It’s unclear what the future holds, but one thing is certain: the economics of AI are changing.
Source: Original article