Over the last two years, the landscape of the artificial intelligence race has been fairly straightforward: the focus has been on larger models, superior benchmarks, and which company can take the lead—until the next product rollout, that is. However, this tally is beginning to appear inadequate.
As organizations transition from simply experimenting with AI to integrating it into tangible products and workflows, the emphasis has shifted. It's not just about selecting the best model anymore; it's about finding the right one tailored for specific tasks, at a feasible cost, with the appropriate data, and in the right environment.
This transformation is paving the way for a new type of AI competition that prioritizes routing, cost efficiency, control, and computational power over sheer model size.
"The model by itself is no longer the sole product," noted Aravind Srinivas, CEO of Perplexity, in an interview with CNBC. "What’s crucial now is the orchestration system that incorporates the model into a robust framework and integrates it with various tools."
Consequently, AI solutions are evolving into comprehensive systems capable of determining the most suitable model to deploy, the optimal timing for its application, and which external tools or proprietary data sources are needed. For instance, a customer service interaction might not require the most expensive model on the market, while a more intricate coding challenge would necessitate it. Alternatively, a straightforward internal process could efficiently operate on a less costly open-source model, whereas more complex tasks could be escalated to a more potent one.
"The optimal choice is always to utilize the best tool for the job," Srinivas emphasized.
This rise of alternative models coincides with a tightening of budgets for AI investments across corporate America, posing an additional challenge to leading firms like OpenAI and Anthropic, which have thrived in recent years by offering the latest technological advancements.
