OpenAI, Meta and SpaceXAI race to cut AI model costs

Three major AI developers released new models within days of each other this week, and all of them lead with the same pitch: lower cost per task. Lorelei Smillie and Rachel Metz report for Bloomberg that OpenAI’s GPT-5.6, SpaceXAI’s Grok 4.5 and Meta’s Muse Spark 1.1 all promise more output for fewer tokens, the units of data AI models process.

The shift reflects growing pressure from business customers. Earlier this year, companies pushed employees toward heavy AI use, a trend nicknamed “tokenmaxxing”. Now many face sticker shock, partly because providers like Anthropic have moved to usage-based pricing. Gautier Cloix, CEO of Paris-based H Company, says he has seen invoices for AI usage reaching millions of dollars a month.

Efficiency becomes the new battleground

Meta CEO Mark Zuckerberg says his company can afford to be “aggressive” on pricing thanks to its advertising business, arguing that competitors’ margins are “very extreme”. OpenAI CEO Sam Altman strikes a similar tone, telling CNBC that enterprises are closely weighing cost against value, a marked change from earlier talk of thousand-dollar subscriptions for top-tier models.

Beyond new models, OpenAI has introduced spending controls and usage analytics to help businesses track costs. Meanwhile, cheaper alternatives keep multiplying: Chinese firms like DeepSeek offer lower-cost open models, and routing services such as OpenRouter, which raised over $100 million in May, let users pick the most affordable model for each task.

Analysts see this cost focus as a way to pressure Anthropic, whose Opus and Fable models rank among the priciest per task, according to Artificial Analysis. Musk directly targeted the rival this week, calling Grok 4.5 “an Opus-class model, but faster, more token-efficient and lower cost”.