Ramp's Data Shows OpenAI's Token Prices Fell 35% in a Month While Its Share of Business Spend Shrank

If OpenAI's price cuts were meant to win business spending, Ramp's latest data does not show it working yet. On Ramp's platform, the blended price businesses paid for OpenAI tokens fell 34.7% in a month, to $0.44 per million, yet OpenAI's share of token spending fell from 44.2% to 35.1%. Anthropic, whose blended price fell less, by 17.9% to $0.73, took 60.5% of spending, up from 51.3%.

The numbers#

Ramp's AI Index puts the blended price per million tokens across OpenAI and Anthropic at $0.59 in the week ending October 23, 2026, against $0.77 in the week ending September 25, a 24.1% drop. Over the same comparison, total token volume rose 37.1% and total token spending rose 8.9%. By volume rather than spend, Anthropic's share rose to 59.8% from 56.1% and OpenAI's fell to 37.2% from 40.2%.

Axios reported the index on Wednesday alongside OpenAI's launch of GPT-6.1 Terra, which OpenAI prices at $1 per million input tokens and $5 per million output tokens. Axios said the index was about 59 cents per million tokens on October 23, roughly 35% below its level at the end of June, and that recent Ramp figures suggest spending growth may be flattening.

What Ramp says is driving it#

In an October 13 newsletter, Ramp economist Ara Kharazian said: "This decline in spend is driven almost exclusively by competition between OpenAI and Anthropic," noting that open-source models make up less than 5% of business spend on Ramp. The same newsletter said businesses are changing AI model providers more than ever.

A blended price falls when list prices drop but also when customers shift toward cheaper models or cached tokens, and Ramp's page does not separate those effects. The month-on-month figures alone cannot say which one is at work.

What the numbers add up to#

Two things stand out. First, most of the price cut went to customers rather than to the labs: volume rose 37.1% but spending only 8.9%, so businesses bought far more tokens for a little more money. Second, Anthropic's lead on this measure is wide. In May, Ramp's adoption series, which counts the share of businesses paying for each provider, showed Anthropic edging past OpenAI for the first time, at 34.4% against 32.3% in April. The token series measures something different, spending by the businesses that run tokens through Ramp's product, and there Anthropic now takes about six dollars in ten despite charging more per token.

That pattern fits the newsletter's account of competition between the two labs, and it suggests price alone is not deciding where the work goes. The data cannot show what is, whether model quality on coding work, existing contracts or something else.

Limits#

• The token figures come from businesses using Ramp's Token Spend Management product, which Ramp says "is not representative of Ramp's total AI spend volume, though results are directionally similar."
• The adoption and token series measure different things and cover different sets of businesses, so they cannot be compared point for point.
• The comparison is one week against the same week a month earlier, so it can swing with a single large customer or launch.
• Axios's "roughly 35% since the end of June" figure is not shown on Ramp's public index page, which gives only the month-on-month comparison.

Sources#

• Ramp Economics Lab: Ramp AI Index (data through September 2026; Tokenomics for the week ending October 23, 2026)
• Ramp Leading Indicators: Token pricing power falls ahead of anticipated mega-AIPOs (October 13, 2026)
• Axios: OpenAI's new model cuts prices again (October 28, 2026)
• Times of AI: OpenAI Prices GPT-6.1 Terra at $1 and $5 per Million Tokens (October 27, 2026)