AI race shifts from large models to cheaper, smarter systems
Artificial intelligence companies are moving beyond a race for the largest models to developing systems that intelligently select the most cost-effective model for each task. The focus is shifting to routing, cost, and integration over sheer model size.
The intense competition in artificial intelligence is evolving beyond a simple race to develop the biggest and newest models. Companies are now shifting their focus towards creating more efficient, economical, and intelligent systems that can better serve specific needs.
Aravind Srinivas, CEO of Perplexity, stated that the model itself is no longer the primary product. Instead, the encompassing "harness" or orchestration system that integrates the AI model with various tools and data sources is becoming key. These systems can autonomously decide which model to employ, when to use it, and what external resources are required, optimizing for task efficiency and cost.
This strategic shift coincides with a broader trend of corporations tightening their AI budgets. The increasing capability and decreasing cost of open-weight models present a significant challenge to the business models of large AI labs like OpenAI and Anthropic. Peter Fenton, a partner at Benchmark, predicts that open-weight models could soon account for over 90% of generated tokens, putting pressure on the profit margins of frontier model providers.
Furthermore, the ease of deploying and managing open models is improving, with companies like Ollama enabling businesses to adopt these solutions. Smaller, task-specific models can often outperform larger, general-purpose models, offering both speed and efficiency. This trend is making AI more accessible and cost-effective for a wider range of applications across industries.