Top AI developers must now find ways to maximize value for current and potential customers, without undercutting themselves to the point where they’d struggle to recoup the hundreds of billions of dollars they’ve invested in chips and data centers.
AI Giants Shift Battle From Raw Power to Price as Enterprise Bills Pile Up
OpenAI, Meta, and SpaceXAI have all released new models within days of each other this month, and for the first time in the AI race, the headline feature isn’t raw capability — it’s cost. All three companies are now marketing token efficiency as their key differentiator, a direct response to enterprise customers who are increasingly balking at ballooning AI bills.
The new releases
SpaceXAI’s Grok 4.5, launched July 8, was pitched by Elon Musk as an Opus-class model that runs faster and cheaper than comparable offerings. Early benchmark data suggests the model completed coding tasks using a fraction of the tokens required by Claude Opus 4.8, and it’s priced at roughly $2 per million input tokens and $6 per million output tokens — more than 60% below Claude Opus and GPT-5.5 rates.
OpenAI’s GPT-5.6, released July 9 in three tiers, is built around the same efficiency pitch. CEO Sam Altman said the flagship model delivers over 50% better token efficiency on agentic coding tasks compared to its predecessor, letting customers get more done per dollar spent.
Meta’s Muse Spark 1.1 rounds out the trio. CEO Mark Zuckerberg told Bloomberg the company is pricing the model aggressively, leaning on Meta’s advertising revenue base to undercut rivals on cost rather than compete purely on benchmark scores.
Why the sudden focus on price
The shift follows a period industry insiders have dubbed “tokenmaxxing” — when companies actively encouraged employees to use AI as heavily as possible with little regard for cost. That era appears to be ending. Enterprises that received unexpectedly large invoices have started imposing usage limits, and providers are responding in kind.
Altman told CNBC that every enterprise today is scrutinizing what it spends on AI against the value it gets back — a striking change in tone from roughly a year ago, when OpenAI executives floated the idea of premium subscriptions costing thousands of dollars a month for top-tier models.
Where this leaves Anthropic
Anthropic finds itself under the most direct pressure from this wave of price cuts. The company recently moved Claude Enterprise from flat-rate subscription pricing to usage-based billing, a shift some read as a sign that rising compute costs are being passed through to customers rather than absorbed.
That’s a notable position given Anthropic’s enterprise subscription share reportedly overtook OpenAI’s for the first time in May. But with three major rivals now competing aggressively on price — and cheaper open-weight models from developers like DeepSeek adding pressure from below — the strategic question for Anthropic will be whether it can match this efficiency push or continue to compete primarily on capability.
The bigger picture
Model-routing platforms that let businesses automatically switch between providers based on cost and task are also gaining traction, with one, OpenRouter, raising over $100 million in fresh funding in May. Together, these signals point to a maturing AI market where the deciding factor for enterprise buyers is shifting from “which model is smartest” to “which model gets the job done for the least money.”