OpenAI said GPT-5.6 Sol is 54% more efficient on agentic coding tasks and may face rollout capacity constraints, while xAI touted Grok 4.5 token efficiency as enterprises focus on AI ROI.
OpenAI and xAI are increasingly competing on enterprise AI pricing and token efficiency as corporate buyers shift from experimentation to tighter scrutiny of usage costs and return on investment. OpenAI introduced GPT-5.6 Sol, part of a new model family that also includes Terra and Luna, and said the model is 54% more efficient on agentic or genetic coding tasks while performing as well as or better than leading rivals. Sam Altman said businesses are now focused on spend, efficiency and ROI, warned of possible "hiccups and capacity crunches" during rollout, and said a broader release was planned for July 9, 2026 after limited partner testing and a U.S. government safety review. xAI has made a similar pitch for Grok 4.5, saying it uses tokens twice as efficiently as rival products. The broader push reflects a market in which lower token consumption can materially reduce the cost of coding, research and support tasks run at scale.