The Data That Broke the Narrative
Over the past 30 days, Ramp's enterprise spend data reveals a number that should unsettle every AI infrastructure investor: Anthropic's flagship model, Fable 5, commands only 11.4% of the company's API expenditure share and a paltry 6% of token volume.
Let that sink in.
The most capable model Anthropic has ever shipped—priced at $10 per million input tokens and $50 per million output tokens, double the cost of its sibling Opus 5—is being rejected by the very enterprises it was designed to serve. Meanwhile, OpenAI's GPT-5.6 Sol captures 25% of enterprise token usage. The asymmetry is not marginal. It is structural. And it tells us something fundamental about where the AI market is heading: we have entered the era of diminishing returns on frontier intelligence.
In DeFi, liquidity is the only truth that matters. In AI, adoption is the only truth that matters.
The Context: A Tale of Two Models
Anthropic positioned Fable 5 as its technical crown jewel. The pricing structure was aggressive: $10/M input, $50/M output. For context, that's 5x the cost of GPT-4-class models and 2x the cost of Opus 5. The company's own messaging framed Opus 5 as "near-frontier intelligence at half the price"—an admission that the capability delta between the two models is not proportional to the price delta.
What the data actually reveals is a market in the process of sophisticated recalibration. Enterprises are no longer chasing benchmark supremacy. They're building AI-driven workflows that require predictable costs, not SOTA scores.
Accel partner Miles Clements said it plainly: "Most people don't need frontier models on a continuous basis." The Ramp data validates this thesis with hard numbers.
Here's the real, uncomfortable picture: the average enterprise is discovering that 90% of their use cases are adequately served by mid-tier models. This isn't a technology gap. It's an economics gap. When you're running 1 million API calls per day, a doubling in unit price translates to millions of dollars in annual infrastructure spend. And the marginal quality improvement from Fable 5 simply doesn't justify that line item.
The Core: Order Flow, Tokenomics, and the Hidden Architecture
The token-to-spend ratio reveals the market's true positioning.
Fable 5's token usage share sits at 6%, while its spend share sits at 11.4%. This disparity—roughly a 1.9x ratio between spend and usage—is the most interesting data point in this entire analysis. It tells us that Fable 5 is being used for high-value, low-frequency tasks: complex reasoning, long-chain planning, sophisticated code generation. The model is not being used for high-frequency, cost-sensitive applications. Enterprises are using it as a scalpel, not a hammer. That's an intentional, rational strategy.
But here's the counterintuitive part: what if Anthropic priced Fable 5 this way deliberately? What if this isn't a failure of strategy, but the successful execution of one that the market hasn't fully parsed?
Anthropic's product architecture now resembles a classic decoy effect framework. Fable 5's pricing serves as the anchoring point against which Opus 5's "near-frontier capability at half the cost" looks attractive. From this perspective, the flagship's "failure" is actually the most effective sales mechanism for the mid-tier model. Every sales call that starts with Fable 5's price and ends with Opus 5's alternative is a conversion that was structured by design.
That's not a failure. That's a route, and I suspect it's working.
From my own experience analyzing tokenomics and yield dynamics in DeFi, I've learned that pricing structures are never arbitrary. They're designed around specific behavioral incentives. The Fable 5/Opus 5 structure is a textbook example of creating a value gradient that pushes spend toward the intended target. This mirrors what I've seen in cryptocurrency markets: the highest-valued assets are often the least traded, serving as store-of-value anchors while the actual transactional volume flows through secondary tokens.
The Contrarian Angle
The lazy take: Anthropic's flagship is failing. The smart money take: the enterprise AI market is being repriced in real time, and we should be watching the data streams, not the benchmarks.
What the Ramp data doesn't capture is the wholesale side of the enterprise equation. My experience with large institutional contracts tells me that the API spend patterns visible on platforms like Ramp skew toward mid-market and SMB companies. Large financial institutions, healthcare providers, and government contractors who need dedicated capacity, private deployment, or contractual SLAs are far less likely to be processing Fable 5 API calls through a spend-management platform. The data might be masking a bifurcation: public API consumption that looks tepid, private/enterprise consumption that's entirely invisible.
There's also the question of safety and compliance as a differentiator. Anthropic's brand has been built on its "safety-first" positioning. In the era of the EU AI Act and mounting regulatory scrutiny, enterprises in regulated industries may be opting for Anthropic models for risk reasons, not performance reasons. The cost of non-compliance in regulated industries is dramatically higher than the cost of a model underperforming by a few percentage points. If Anthropic's models are being selected for safety, they may be deployed in a more conservative manner—fewer tokens, but higher-value tasks—which would explain the token/spend disparity.
This is the blind spot in the market's read on the data: enterprises may be selecting Anthropic for regulatory risk minimization, not performance maximization. And they may be doing it through channels that Ramp doesn't track.
The Industrial Ripple: What This Means for the Entire AI Supply Chain
The enterprise procurement signal is moving from "best model" to "best value." That transformation will ripple through the entire AI ecosystem.
Model vendors. Anthropic's pricing architecture needs to adjust to a world where a company's CTO is more likely to face procurement scrutiny. The "build a better model and they will come" era is over. The era of "build a model portfolio that addresses segmented demand" has begun.
OpenAI's advantage: GPT-5.6 Sol's 25% token share suggests that OpenAI's pricing structure more closely aligns with enterprise demand curves. Whether it's volume discounts, bundled contracts, or more flexible tiering, OpenAI is capturing the "volume" segment. If OpenAI also moves into the high-end with an aggressive price, the pressure on Anthropic intensifies.
Infrastructure providers: The lower-than-expected demand for flagship-class inference has implications for GPU utilization. A significant portion of high-end GPU demand was built on the assumption that every token generated would be the most expensive token. That assumption is now in doubt. Data center operators who over-provisioned for frontier inference may face idle capacity.
Application layers: If enterprises start routing their tasks to mid-tier models, the entire AI middleware space shifts. The next unicorn is already being built on model routing—intelligent middleware that automatically routes tasks to the cheapest model capable of handling them. Enterprises will adopt this for cost optimization and performance optimization.
The procurement logic has shifted to the fundamentals: price, reliability, and ROI.
Where the Market Is Headed
The data tells us one thing clearly: enterprise AI is in the early stages of a new cycle. The market is moving from a "technology push" phase to a "demand pull" phase.
The next phase of the model market will not be about benchmark dominance. It will be about a pricing architecture, an ecosystem, a safety case, and an ability to build into existing workflows. The winners will be those who can deliver 90% of the capability at 30% of the cost, and make that positioning understood. The losers will be those who cling to the "frontier premium" narrative.
Greed is a variable; discipline is the constant.
The Bottom Line: The Real Battle Is Not About Performance
The quiet truth about Anthropic's Fable 5's low adoption rate is that it represents a structural repricing of AI value. The market is telling us that the marginal utility of the frontier model has peaked. The era of blind "best-model" procurement is over.
The question now is whether Anthropic will adapt quickly enough to the new market reality, or whether it will continue to pour resources into a product that the market has clearly told it is overpriced. The next 6-12 months of enterprise procurement will be the ultimate test.
If Anthropic can pivot to making Opus 5 the primary growth engine and leverage Fable 5 as a strategic anchor, it can thrive. If it treats the Fable 5 data as a data issue rather than a strategic signal, the long-term cost will be measured in market share.
In this market, the truth is the data. The data is brutal. The enterprise has spoken. The question is who will listen.