For more than two years, artificial intelligence companies competed to build the most capable models, but the battle is now shifting to value for money as businesses face soaring operating costs from AI's vast computing power. OpenAI’s newly launched GPT-5.6 family is designed to use significantly fewer tokens, lowering computing costs. Meta’s new Muse Spark 1.1 model, aimed at developers, will be priced "very aggressively," with introductory credits before usage-based pricing that undercuts rivals. Elon Musk’s xAI says Grok 4.5 delivers roughly twice the token efficiency of comparable models. Technology companies have collectively committed hundreds of billions of dollars to AI infrastructure, but enterprise customers increasingly expect lower operating expenses. Anthropic executives argued that cutting AI altogether is "the wrong" response, urging businesses to optimise model use instead. Industry observers note companies are becoming more disciplined about AI spending, placing greater value on practical returns over marginally smarter models.