According to a report by TechNews, OpenAI, SpaceXAI (formerly xAI) and Meta are racing to cut per-token costs with new models, while OpenAI CEO Sam Altman says enterprises are scrutinizing AI spending — a shift also flagged by Cnyes as 'AI bill anxiety' driving vendors to compete on token value.
How are AI model makers using technical optimization to cut token costs?
According to a report by TechNews, OpenAI said its most advanced model, GPT-5.6, is "designed to complete more work while dramatically reducing token usage," which the company says makes it more cost-efficient for customers 1. Separately, SpaceXAI — the company formerly known as xAI — launched Grok 4.5 and claimed its token efficiency is "2x that of other companies' comparable models," per the same TechNews report 2. Both companies are framing token efficiency, rather than raw capability, as the headline feature of their newest releases.
How do pricing strategies differ across vendors?
Meta CEO Mark Zuckerberg said Meta will offer "an extremely attractive price" for its newest Muse Spark 1.1 model, according to TechNews 3. In comments to Bloomberg cited by TechNews, Zuckerberg went further, saying "other AI labs' pricing is very high, and their margins are quite astonishing. We think there's a real opportunity to offer state-of-the-art AI capability at a much more affordable price" 4. On the same day, Elon Musk promoted Grok 4.5 in a post directly targeting Anthropic, stating: "Grok 4.5 is a Claude Opus-class model, but faster, more token-efficient, and cheaper," as reported by TechNews 9. Taken together, three separate companies — Meta, SpaceXAI and, by implication, OpenAI — are each publicly positioning their newest model against rivals specifically on price and token efficiency rather than on benchmark performance alone.
What cost pressures are enterprise customers facing, and how are procurement decisions changing?
OpenAI CEO Sam Altman told CNBC, as reported by TechNews, that "every company right now is thinking about AI spend, and how much value that spend is actually generating — that's exactly what we want to enable" 5. This corporate scrutiny is echoed in a Cnyes headline describing the phenomenon as "AI bill anxiety," reporting that OpenAI, Meta and Musk have all begun competing on "token value for money" 10. The two outlets frame the same underlying dynamic from different angles: TechNews via Altman's direct quote about enterprise value-tracking, and Cnyes via a framing that ties enterprise anxiety directly to the vendors' competitive response.
How is OpenAI upgrading tools to help customers manage spending?
According to TechNews, OpenAI has also started helping enterprises manage AI spending directly: last month, it introduced credit usage analytics for ChatGPT Enterprise and updated its spending-control mechanisms 6. This positions OpenAI's product changes as a direct response to the enterprise cost-scrutiny that Altman himself described 5, pairing a stated market observation with a concrete product action.
What does the current cost-efficiency landscape look like across models?
Not every vendor is repositioning on price. Data from AI benchmarking service Artificial Analysis, cited by TechNews, shows that Anthropic's Claude Opus and Claude Fable models remain, on a per-task cost basis, among the most expensive models on the market 8. This is the specific gap that Musk's Grok 4.5 pitch 9 and Zuckerberg's high-margin critique 4 appear to be targeting — both comments reference the more expensive end of the market that the Artificial Analysis data identifies.
What financing and business opportunities has the cost-efficiency race created?
The scramble for token efficiency has also created a market for intermediary services. TechNews reports that OpenRouter, a company offering model-routing services that let customers pick among models for cost or performance, completed a funding round of more than $100 million in May, which the report frames as a signal that investors see strong prospects for this kind of service 7.
What this means
The evidence points to a consistent pattern across three separate vendors — OpenAI, SpaceXAI and Meta — each publicly emphasizing token efficiency or price in their most recent model launches (E1, E2, E3, E4, E9), at the same time that OpenAI's own CEO and a Cnyes report both describe enterprise customers actively scrutinizing AI spend (E5, E10). OpenAI's decision to add usage analytics and spending controls to ChatGPT Enterprise 6 lines up directly with that scrutiny. Meanwhile, Artificial Analysis data showing Anthropic's Claude Opus and Claude Fable as the most expensive models on a per-task basis 8 gives concrete grounding to the price comparisons Musk and Zuckerberg are making in public statements (E9, E4), and the $100 million-plus funding round for model-router OpenRouter 7 suggests investors are betting that customers will keep needing tools to navigate the resulting price differences.