Ramp Data Suggest a Price Ceiling for Anthropic’s Flagship AI Model

Ramp’s July 2026 AI Index points to a widening gap between enterprise adoption and premium-model demand. Anthropic remained slightly ahead of OpenAI in the share of tracked businesses paying for its products, but its most expensive flagship model, Fable 5, attracted far less usage than OpenAI’s GPT-5.6 Sol. 2

That does not establish that Anthropic’s companywide prices, revenue growth, or profitability have fallen. Ramp’s data measure spending and token activity among more than 70,000 businesses using its financial platform—not Anthropic’s consolidated income statement. The stronger conclusion is narrower: businesses may be willing to buy AI, while remaining reluctant to pay the highest available frontier-model prices.

Fable 5’s first-month adoption was modest

In its first month, Fable 5 accounted for only 6% of tokens purchased from Anthropic and 11.4% of spending on Anthropic models in Ramp’s sample. Ramp described the sample as tech-skewed and cautioned that the model’s true adoption could be lower. 2

OpenAI’s GPT-5.6 Sol showed stronger pull within OpenAI’s product mix, representing 25% of tokens and 23% of spending. Fable 5 generated approximately 75% as much model-attributed business spend as Sol during July. 2

The comparison is significant because Fable 5 was positioned as Anthropic’s flagship model, while lower-priced alternatives were available. OpenAI lists GPT-5.6 Sol API pricing at $5 per million input tokens and $30 per million output tokens. 6 The available Ramp evidence therefore supports a case for enterprise price sensitivity, although it does not by itself prove that price was the only reason for the adoption gap.

The broader competition is cheaper models, not just OpenAI

Enterprises can choose among several layers of AI infrastructure: premium proprietary models, cheaper models from the same provider, open-source systems, and models developed in China. Ramp’s index reported that model-serving platforms offering open-source and Chinese-developed models reached 6.1% of AI-using firms in its measure. 2

That availability changes the purchasing decision. A company does not need the single most capable model for every workflow. It can reserve an expensive model for difficult reasoning or high-value tasks and use less costly systems for routine classification, drafting, extraction, coding assistance, or customer-service workloads.

The result is a two-part market: frontier models compete on peak capability, while day-to-day enterprise volume is governed by cost, latency, reliability, data policies, and measurable business outcomes. A small performance advantage may justify a premium in a critical workflow, but not across every request.

Anthropic still led in business reach

The Fable 5 result should not be confused with an overall loss of enterprise adoption. In July, 43.5% of U.S. businesses tracked by Ramp paid for Anthropic subscriptions or tokens, compared with 39.7% for OpenAI. 2

Anthropic’s share increased by 1.1 percentage points from the prior month, while OpenAI’s rose by 0.23 percentage points in the same comparison. 11 Later Ramp-related reporting indicated that OpenAI was expanding faster than Anthropic among U.S. business users in the third quarter to date, even though Anthropic retained the July lead. 3

The most defensible reading is therefore: Anthropic had broader reach in this sample, while OpenAI’s flagship captured stronger relative demand and OpenAI’s business-user growth was accelerating. Ramp’s sample is useful but not a complete census of the enterprise AI market, so the figures should be treated as directional rather than as total market share.

What the data do—and do not—say about Anthropic’s finances

Ramp’s index cannot establish Anthropic’s average realized model price, total revenue, gross margin, operating margin, or profitability. Token mix and card or bill-pay activity are not substitutes for audited company financials.

The data also do not show that Anthropic’s overall revenue growth has stopped. They show that one flagship model generated less business spend than a competing flagship during the month observed. Anthropic could still grow rapidly through subscriptions, coding products, enterprise contracts, lower-priced models, higher usage of other Claude models, or broader customer adoption. Those possibilities are not resolved by the Ramp model-level comparison.

This distinction matters for investors. A company can grow revenue quickly while the price of its most advanced model comes under pressure. Growth may then depend increasingly on volume, product breadth, workflow integration, and customer retention rather than on charging a large premium for every frontier-model interaction.

Why this matters for a potential $2 trillion IPO

Reuters reported that Anthropic was projecting roughly $190 billion to $200 billion in 2028 revenue, figures being considered in discussions around a potential valuation near $2 trillion. 17

Ramp’s data do not invalidate that forecast, but they highlight the execution burden behind it. If enterprises limit their use of the most expensive models, Anthropic must reach those revenue targets through some combination of:

• much higher model usage;
• large-scale adoption of coding and workflow products;
• expanded enterprise distribution;
• lower-cost models that generate high request volumes; and
• durable performance advantages in workflows where accuracy has measurable economic value.

A flagship launch alone is not enough. The commercial test is whether customers use the model repeatedly, at scale, and at prices that support attractive economics.

The pricing lesson for the AI industry

The evidence suggests that enterprise willingness to spend on AI is real but selective. Ramp reported median AI spending of $7,400 per employee among the top 1% of firms by AI expenditure in July. 7 High budgets, however, do not mean every company will choose the most expensive model for every task.

AI providers are consequently likely to compete with more differentiated pricing structures: premium reasoning models for complex work, inexpensive models for high-volume workloads, enterprise bundles, usage tiers, and pricing tied more closely to business outcomes. Providers may also need to make the return on investment clear enough for procurement teams to justify moving beyond experimentation.

For Anthropic, Fable 5’s early numbers are best understood as a warning about premium monetization. The company still had a business-adoption lead in Ramp’s July sample, but its flagship model did not command a proportionate share of enterprise usage. That is a manageable product and pricing problem—yet one that becomes more consequential when future valuation expectations depend on extraordinary revenue growth. 2