EnterpriseSEP 13, 2026

Token prices fall 41% to $0.68 as top 1% of AI spenders cut per-employee outlays 9.7%

Ramp's September AI Index shows effective per-million-token prices down from a $1.15 March peak, with frontier model share slipping to 45% as businesses steer toward mid-tier Terra and Sonnet.

BY GRETA REINHART · ENTERPRISE SAAS

The effective price of a million AI tokens has fallen 41% from its March peak of $1.15 to $0.68, according to the September Ramp AI Index published Sept. 9. The same report shows the top 1% of AI spenders cutting median per-employee spend 9.7% in August, from $7,976 to $7,205. Two data points, one story: the customers who actually know what they’re paying for are paying less and buying cheaper tiers.

Frontier model share of token volume slid from a 53% peak in early August to 45% by September, with usage migrating to mid-tier options. Ramp lead economist Ara Kharazian told Fortune that businesses are “imposing defaults that steer employees away from frontier models entirely,” describing GPT-5.6 Terra and Claude Sonnet as “highly performant and also cheaper.” Opus, Fable, and Sol, the reigning frontier tier, are losing share not because they got worse but because the rung below caught up.

Provider math tells the same story from the sell side. Fortune reports OpenAI’s effective price is down 38% since Aug. 1 to $0.48, Anthropic’s down 22% to $0.90. At a Goldman Sachs conference, OpenAI CFO Sarah Friar said the company has cut GPT-5.6 Luna’s price 80% since launch and added, “I would love to get us away from token-counting.” CNBC noted Silicon Data’s LLM Token Expenditure Index fell to 97 cents on Monday, the lowest reading since the index’s creation late last year.

The through-line links back to OpenAI’s reset of the flagship price curve last week and, upstream, to Nvidia’s $96.2B quarter: supply arrived, and pricing power at the top of the stack is compressing accordingly.

For a 5-to-30-person company, compute costs for AI-driven workflows are materially lower than they were two quarters ago. The awkward part is that most small firms aren’t running the workflows: Ramp’s August index put the median firm at $11.95 per employee for July. The wait-for-prices-to-drop thesis has largely paid out. What’s left to gate the budget is the pending Fed rate decision, which will decide whether cheaper tokens actually translate into more software line items or simply cheaper ones.

Sources

About the author

Greta Reinhart

ENTERPRISE SAAS

Greta Reinhart tracks the enterprise software stack from San Francisco — data platforms, AI bundling, seat pricing, and channel checks across the largest SaaS vendors. She files on go-to-market shifts, packaging changes, and quarterly enterprise reads.