The artificial intelligence industry is moving beyond the simple metric of model size as companies prioritize cost efficiency and task-specific optimization over raw capability.

Perplexity CEO Aravind Srinivas says the real product is becoming the orchestration system that selects which model to use for each task. "The model alone is no longer the product," Srinivas told CNBC. "It is the harness, the orchestration system that puts the model inside a very capable harness and pairs the model with a lot of tools."

This shift reflects how companies are deploying AI in production. A customer service query might run on a cheaper model, while complex coding problems get routed to more powerful systems. Routine workflows could use open models, with difficult tasks escalated to premium alternatives.

Perplexity this week previewed a computer-use system built around GLM 5.2, an open model from China's Z.ai. The system uses cheaper models for most work while calling stronger models only when necessary.

Open models challenge proprietary leaders

Benchmark general partner Peter Fenton predicts 90% of AI tokens will come from open-weight models within 18 to 24 months. "The inference margins generated by the frontier model companies are going to come under pressure when you can run those without the markup they're providing," Fenton said.

Open-weight models can be downloaded, customized, and run by companies directly, often at lower cost than proprietary alternatives from OpenAI and Anthropic. Benchmark invested in Ollama, which helps enterprises deploy open models locally.

Ollama CEO Jeff Morgan says the company has been adopted by more than 85% of Fortune 500 companies, including regulated industries like aviation and healthcare. Many start with smaller local models before expanding to larger open alternatives.

The trend creates strategic challenges for the US, as many competitive open models originate from Chinese labs including Z.ai and DeepSeek. Srinivas argues America should support open models to make AI more affordable for small businesses.

The shift could also affect data center construction, with some AI work potentially moving to local devices rather than centralized cloud infrastructure.