The AI race is shifting from bigger models to cheaper, smarter systems

The AI race is shifting from building larger models to deploying cheaper, smarter systems designed for specific tasks and cost control, according to Benchmark’s Peter Fenton and Perplexity CEO Aravind Srinivas. Srinivas told CNBC that “the model alone is no longer the product,” arguing that the differentiator is the orchestration and harness that routes requests to the right model and tools. The report says companies are moving from pilots to real product workflows, which increases emphasis on routing, compute, control, and the quality of data access rather than leaderboard benchmarks. It cites Perplexity’s preview of a computer-use system built around GLM 5.2, an open model from China’s Z.ai, aiming to use a cheaper model for most work while escalating only when needed. Fenton said open-weight models could generate 90%+ of tokens in 18–24 months, putting pressure on frontier-model providers’ inference margins. The article also notes that OpenAI and Anthropic face added challenges as corporate AI spending tightens.