IBM CEO says businesses are choosing open-weight models over expensive frontier AI

IBM CEO Arvind Krishna said during IBM’s third-quarter 2026 earnings call that enterprises are increasingly selecting open-weight AI models because they can deliver roughly 90% of frontier-model performance at 10% to 25% of the token cost. He said businesses have become more focused on return on investment and are concerned about AI spending that produces no clear return. Krishna said the performance gap between leading frontier models and alternatives has narrowed sharply since 2024. He characterized the earlier gap as benchmark scores of 70 for the top model versus 20 or 30 for open models, improving to about 70 versus 50 in late 2025, with a potential future comparison of 100 versus 90. He said some tasks still benefit from higher-cost frontier systems, but that lower-cost open-weight models can handle most enterprise tasks. Krishna also identified AI sovereignty as a key reason companies deploy open-weight models on their own infrastructure or private clouds. He said organizations handling sensitive data want control over logs, data and AI integrations rather than leaving traces with public LLM providers. He said a large enterprise can begin operating open-weight models with two or three racks of infrastructure.