OpenAI is gaining on Anthropic in Ramp's data, and the lever was price, not capability

Ramp's model-level percentages can be worked backwards into something the release does not state: the relative size of the two books. Fable 5 took 11.4% of Anthropic's model dollars ^[5] while generating about 75% of what GPT-5.6 Sol generated ^[7], and Sol accounted for 23% of OpenAI's model dollars ^[6]. That implies Anthropic's total model-attributed spend across Ramp's businesses runs about 1.5 times OpenAI's ^[17]. Apply the same arithmetic to tokens, using the two-to-one price gap Ramp cites ^[8], and Anthropic is ahead there too, by roughly 1.6 times ^[18]. Both ratios sit near the reported second-quarter revenue gap of $11.6 billion to $6.7 billion, or 1.7 to 1 ^[13][19].

So the leader on Ramp is losing at the margin, not at the level. And the margin moved for reasons that have little to do with which model is better. Ramp economist Ara Kharazian credits Sol's standing with developers, and attributes Fable 5's weak uptake to price plus data retention requirements imposed by regulators ^[3]. Fable 5 lists at about $10 per million tokens, twice Sol ^[8], and Anthropic drew complaints after telling Fable users it must hold their data for a month, according to TechCrunch ^[9]. A price and a retention clause are both things a vendor can change. A capability gap is not.

The catch-up carries a bill on OpenAI's side. A $12.3 billion second-quarter operating loss on $6.7 billion of revenue ^[13][14] implies roughly $19 billion of operating cost, or $2.84 spent for every dollar booked ^[16]. Anthropic recorded a small adjusted operating profit over the same period, according to SiliconAngle ^[14]. OpenAI's own account of the turn is internal and unaudited: CFO Sarah Friar told employees that July annualised recurring revenue had already passed the full second-quarter total, crediting the GPT-5.6 series, an enterprise agent called ChatGPT Work, and Codex ^[15].

There is one more limit on what this data can bear. Ramp sees only companies that run expenses through Ramp, which leaves out large enterprises on other spend-management platforms ^[12]. What it measures well is mid-market buying behaviour, and that behaviour is getting less loyal rather than more: paid AI adoption reached nearly 56% of tracked businesses in July from just over 50% in March, with momentum behind both vendors cooling as buyers try cheaper and open-source options ^[10]. A book that reprices this fast is not a vendor bet in any useful sense of the word.