Will tech companies switch to cheaper AI models

Until now, the AI industry has grown rapidly following Scaling Law, under the assumption that bigger models are more powerful. However, as the cost of operating models has recently risen, a Cost-conscious model-shopping trend is emerging in which companies seek cheaper models without sacrificing performance.

Coinbase co-founder Brian Armstrong predicted that within the next 12-18 months, 80% of all workloads will shift to models that are 99% cheaper than current ones. Only the remaining 20%, he forecasts, will use the latest models for tasks that require top-tier intelligence.

Indeed, the legal AI tool Harvey partnered with Fireworks AI to mix Claude Opus and GLM 5.1, successfully cutting inference costs by 3x without any quality degradation. This suggests that instead of unconditionally using the most powerful model for every task, it has become important to choose the optimal model suited to the nature of the task.

This shift is likely to be reshaped not simply as competition between closed models and open-weight models, but as a dynamic between Large models and Small models. If it is proven that small models are sufficient for most tasks, it will raise fundamental questions about the massive inference demand for developing giant models and the justification of their training costs.