The AI race shifts from who has the biggest model to who uses it best

For two years, the AI competition was simple: build the largest model, top the benchmarks, and claim the lead until the next release. That scorecard is now outdated. As companies move from testing AI to deploying it in real products and workflows, success no longer depends on having the most powerful model - it's about finding the right model for each job, at the right cost, with the right data, in the right environment.

This change opens the door for a new kind of competition focused on routing, cost, control, and compute, not just model size. Perplexity CEO Aravind Srinivas told CNBC, "The model alone is no longer the product. It is the harness, the orchestration system that puts the model inside a very capable harness and pairs it with a lot of tools." AI systems now decide which model to use and when - a customer service query might run on a cheap local model, while a complex coding task escalates to a premium one. Open-weight models, which companies can download and tune themselves, are becoming cheaper and more capable, putting pressure on premium proprietary labs like OpenAI and Anthropic.

Benchmark general partner Peter Fenton predicts a dramatic shift: "90-plus percent of the tokens created will come out of open-weight models over the next 18 to 24 months." He notes that specialized smaller models can outperform larger general-purpose ones for specific tasks. That's why Benchmark invested in Ollama, a platform that helps enterprises run open models. Ollama CEO Jeff Morgan says more than 85% of the Fortune 500 now use the tool, often starting with small models close to their own data and scaling up as they gain confidence. The rise of open models also creates a strategic challenge for the U.S., since many leading open-weight models come from Chinese labs like Z.ai and DeepSeek. Srinivas argues the U.S. should embrace open models to keep AI affordable for small businesses. The shift may also reshape the data center boom, with some AI work eventually running locally on consumer devices, creating a hybrid system where routine tasks stay on-device and only the hardest jobs go to the cloud.

Source: CNBC