“State-of-the-art” (Sota) artificial intelligence models excel at solving complex Olympiad maths but still struggle with everyday enterprise tasks, according to an executive from a top AI unicorn in the US. David Meyer, senior vice-president of product at US data processing and analysis company Databricks, told the South China Morning Post in a recent interview that the very traits making models state-of-the-art could cause issues in basic office work. For instance, when tasked with identifying an erroneous number on an invoice, a Sota model “will oftentimes fix the mistake” rather than simply extracting the error for downstream correction, he said. The discrepancy extends to other highly technical domains as well. While advanced models such as Anthropic’s Claude were powerful at coding, they could lag in tasks like data engineering compared with models with significantly more specialised training and data in this area, according to Meyer. Data engineering involves transforming datasets at scale and performing cleaning tasks, such as handling null values and zeros. “A single model, no matter how large, can’t be equally good at all things,” he said. To solve these specific complexities more efficiently, Meyer pointed to the use of small open-source models refined with reinforcement learning. This allowed for a specific purpose at a level of training cost “orders of magnitude lower” than Sota models, according to Meyer. This strategy is already being reflected in the company’s own tools. Databricks’ Genie, an AI assistant that translates natural language into data queries and is supported by a system of agents and AI models, provides a window into a wider industry trend. By observing how customers interacted with the platform, Meyer noted a surging preference for smaller models over Sota alternatives. These leaner models, which utilise significantly fewer parameters, are gaining traction, driven by both low cost and low latency. “Small models, just by their nature, are a lot faster [in] time to first token and time to response,” Meyer said. When applications scaled to handle “astonishingly high queries per second”, companies needed inexpensive models for that volume. However, the search for these highly capable, cost-effective models comes with geographic limitations. Meyer observed that although “people are very excited about the Qwen models,” and Chinese open-source models were “amazing in terms of capability, low latency and cost”, regulatory and compliance concerns currently limited their use in enterprise settings. The Qwen model series was developed by Alibaba Cloud, the AI and cloud computing unit of Alibaba Group Holding, which owns the SCMP. This restricted landscape has not dampened the overall rush to integrate AI among enterprises. “A lot of people are afraid of being left behind, so they want to push AI as fast as possible,” said Meyer. He added that public companies were generally more cautious about how AI spending affected their balance sheets, while private companies tended to be more willing to spend. Capitalising on this industry-wide investment, Databricks announced in February that it had completed around US$5 billion in equity financing at a US$134 billion valuation. Following more than 65 per cent year-on-year growth in the final quarter of 2025, the company is now on track to generate US$5.4 billion over the next 12 months.