For the last two years, the AI contest was simple: build a bigger model, top a benchmark, and claim the crown until the next release. That scorecard is now out of date. As companies move from testing AI to embedding it in real products and workflows, the goal is no longer to tap the single best model. Instead, it’s about finding the model that fits a specific task—at the right price, with the right data, and in the right environment. This change is opening a new kind of competition, one focused less on raw size and more on routing, cost control, and compute efficiency.
“The model alone is no longer the product,” said Perplexity CEO Aravind Srinivas. “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.” In practice, that means AI products are becoming smart routers: a customer service query might not need the most expensive model, while a complex coding problem might. A routine internal workflow could run on a cheaper open model, and a harder step could be escalated to a more powerful one. “The answer is always use whatever is the best for the task,” Srinivas added. This shift is especially critical as corporate America tightens its AI spending, presenting a fresh challenge for OpenAI and Anthropic, which have thrived by selling cutting-edge technology at a premium.
Perplexity recently previewed a system for its computer-use product built around GLM 5.2, an open model from China’s Z.ai. The design lets a cheaper model handle most of the work, calling in a stronger model only when needed. This mirrors a broader market trend: open-weight models—which companies can download, tune, and run themselves—are getting more capable and far cheaper to operate than proprietary alternatives. Benchmark Capital partner Peter Fenton predicted a dramatic shift: “A contrarian view that is becoming consensus is that 90-plus percent of the tokens created will come out of open-weight models over the next 18 to 24 months, possibly even by the end of the year.” He added that the inference margins of frontier model companies are under pressure when you can run good enough models without their markup. Smaller, task-tuned models can also be faster and more performant than larger general-purpose ones. That’s why Benchmark invested in Ollama, a company helping developers and enterprises download and manage open models. Ollama CEO Jeff Morgan noted that more than 85% of the Fortune 500 have adopted the tool, often starting with small models near their own data and scaling up as they gain confidence.
The rise of open models also creates a strategic challenge for the U.S., as many competitive open-weight models come from Chinese labs like Z.ai and DeepSeek. This makes open-source AI a business, policy, and national competitiveness issue. “If you want the benefits of AI to be widely distributed to small businesses in America and allied countries, you need AI to be a lot more affordable,” said Srinivas. “Open source is the only way to do that.” The shift could also reshape the massive data center buildout. Srinivas predicts some AI work will eventually run locally on consumer or business devices, creating a hybrid system where routine tasks stay local and the hardest work goes to a cloud model. For investors, the big question is whether labs like OpenAI and Anthropic can maintain their pricing power as open models improve and companies become more selective.
Source: CNBC