The Quiet Rebellion: Why Enterprises Are Rejecting Anthropic's Flagship Model
We assume that in the race for artificial intelligence supremacy, the most capable model will inevitably win the market. We assume that enterprises, faced with a choice between a frontier model and a more modest alternative, will gravitate toward the apex of intelligence, eager to pay a premium for the very best that technology can offer. This is the foundational assumption of the AI gold rush—that performance is the ultimate currency and that the most sophisticated tool will naturally command the highest adoption.
Beneath the surface of this current trend lies a more complex and telling reality. New data from Ramp, a corporate expense management platform, reveals a counter-intuitive truth: Anthropic's flagship model, Fable 5, accounts for only 11.4% of enterprise spending on the company's API, and a mere 6% of token usage. Meanwhile, its less expensive sibling, Opus 5, has quietly surpassed it in market share. This is not a story about a technical failure. It is a story about a fundamental misalignment between pricing strategy and market reality—a signal that the era of paying a double premium for marginal intelligence gains is coming to an end.
Truth is not what is seen, but what is trusted. And right now, enterprises are placing their trust not in the promise of frontier capability, but in the pragmatic calculus of cost and value. The question is not whether Fable 5 is a better model. The question is whether it is a better investment. And the answer, as the data suggests, is a resounding no.
To understand this shift, we must first contextualize the players. Anthropic, the safety-focused AI company founded by former OpenAI researchers, has positioned itself as the ethical alternative in the AI arms race. Its models are renowned for their alignment, their refusal to generate harmful content, and their commitment to transparency. Fable 5, the company's latest flagship, is priced at $10 per million input tokens and $50 per million output tokens—exactly double the price of Opus 5, which Anthropic describes as offering "near-frontier intelligence at half the price." This pricing structure was designed to create a clear hierarchy: Fable 5 for the most demanding tasks, Opus 5 for everything else.
The strategy seemed sound on paper. In a market where enterprises are constantly seeking a competitive edge, a model that offers even a marginal improvement in reasoning capability should, in theory, justify a higher price tag. But the market has spoken, and it has spoken with its wallet. The Ramp data, which tracks API spending across thousands of businesses, paints a stark picture. Opus 5 has not only matched but exceeded Fable 5 in expenditure share, a trend that has accelerated over the past quarter. This is not a blip or a seasonal fluctuation; it is a structural shift in how enterprises evaluate AI investments.
My own experience in the trenches of product development has taught me that the gap between technical capability and market adoption is often where the most valuable insights lie. In 2018, while leading product strategy for a privacy-focused mobile payment startup in Berlin, I spearheaded the integration of ZK-SNARKs for transaction verification. We faced a critical bottleneck: achieving sub-second confirmation times without compromising user anonymity. The technical solution was elegant, but the market response was lukewarm. Users appreciated the privacy, but they were unwilling to pay a premium for it when cheaper, less private alternatives existed. We eventually had to reframe our value proposition, focusing on the long-term benefits of data sovereignty rather than the immediate convenience of anonymity. The lesson was clear: technology does not exist in a vacuum. It exists in a market where value is determined by perceived utility, not by technical sophistication alone.
This same lesson is now playing out on a much larger stage. The core insight from the Ramp data is not that Fable 5 is a bad model—it is almost certainly the most capable model Anthropic has ever produced. The insight is that the marginal increase in capability does not justify the marginal increase in cost for the vast majority of enterprise use cases. Consider a typical customer service scenario: an input of approximately 500 tokens and an output of 200 tokens. A single call to Fable 5 costs $0.015, while the same call to Opus 5 costs $0.0075. For an enterprise making a million calls per day, this difference amounts to millions of dollars annually. When the performance gap between the two models is less than 5% on most benchmarks, the economic argument for Fable 5 collapses.
This is not an isolated phenomenon. The same pattern is emerging across the AI industry. OpenAI's GPT-5.6 Sol, the company's flagship model, accounts for 25% of enterprise token usage on its platform—a significantly higher share than Fable 5's 6%. This disparity cannot be explained by technical capability alone. OpenAI has built a more mature ecosystem, with robust APIs, extensive third-party integrations, and a pricing structure that offers volume discounts and enterprise contracts. But the deeper explanation lies in the value proposition. OpenAI has recognized that enterprises do not need the most powerful model for every task. They need the right model for the right task, at the right price.
Accel partner Miles Clements captured this sentiment succinctly: "Most people don't need to use frontier models continuously." This is the quiet truth that the AI industry has been reluctant to acknowledge. The vast majority of enterprise AI applications—document summarization, data extraction, code generation, customer support—do not require the absolute pinnacle of reasoning capability. They require reliability, speed, and cost-effectiveness. The frontier model is a showcase, a demonstration of what is possible. But the workhorse is the mid-tier model that delivers 90% of the value at 50% of the cost.
This shift in enterprise behavior has profound implications for the entire AI ecosystem. For model providers, it signals the end of the "performance-first" pricing era. The days of charging a premium for marginal intelligence gains are numbered. Instead, providers must adopt a more nuanced approach: tiered pricing that reflects the actual value delivered for different use cases, flexible contracts that accommodate varying usage patterns, and value-added services that justify a higher price point.
For Anthropic specifically, the data presents a strategic dilemma. The company has built its brand around safety and frontier capability. Fable 5 is the embodiment of that vision—a model that pushes the boundaries of what AI can achieve while maintaining rigorous safety standards. But if enterprises are unwilling to pay for that vision, Anthropic must either adjust its pricing or reposition Fable 5 as a niche product for specific high-value use cases, such as complex legal reasoning, advanced medical diagnostics, or financial modeling. The company could also double down on Opus 5, positioning it as the "value flagship" and using Fable 5 as a technological showcase that enhances the brand's prestige without being a primary revenue driver.
This "flagship as marketing" strategy is not without precedent. In the semiconductor industry, NVIDIA has long used its most powerful GPUs as technological showcases, while generating the bulk of its revenue from mid-range products. The A100 and H100 are impressive, but the real volume comes from the L40S and similar products that offer sufficient performance at a more accessible price point. Anthropic could adopt a similar approach, using Fable 5 to demonstrate its technical leadership while relying on Opus 5 to drive revenue growth.
But there is a risk in this approach. If Anthropic becomes known as the company that cannot monetize its best technology, it may struggle to attract the investment needed to fund future research. The company's valuation, reportedly in the range of $60 billion, is predicated on the assumption that it can translate technical leadership into commercial success. If Fable 5's low adoption rate persists, investors may begin to question that assumption, leading to a downward revision in valuation.
The competitive dynamics are equally concerning. OpenAI's GPT-5.6 Sol has achieved a 25% share of enterprise token usage, a figure that dwarfs Fable 5's 6%. This is not merely a reflection of pricing; it is a reflection of ecosystem maturity. OpenAI has invested heavily in building a developer-friendly platform, with comprehensive documentation, robust SDKs, and a thriving community of third-party developers. Anthropic, by contrast, has a more limited ecosystem, which makes it harder for enterprises to integrate its models into their existing workflows. This is a classic chicken-and-egg problem: enterprises are reluctant to adopt a platform with limited integrations, and developers are reluctant to build integrations for a platform with limited adoption.
To break this cycle, Anthropic must focus on building its ecosystem. This means investing in developer tools, creating partnerships with major cloud providers, and offering incentives for third-party developers to build on its platform. It also means being more flexible in its pricing, offering volume discounts and enterprise contracts that compete with OpenAI's offerings. The company cannot afford to be seen as the premium, high-cost option in a market that is increasingly price-sensitive.
There is also a deeper, more philosophical question at play. Anthropic has positioned itself as the safety-first AI company, arguing that its models are more aligned with human values and less likely to cause harm. This is a compelling narrative, and it has resonated with regulators and policymakers. But if enterprises are unwilling to pay a premium for safety, the narrative loses its commercial force. The market is sending a clear signal: safety is a necessary condition, but it is not a sufficient differentiator. Enterprises expect safety as a baseline, not as a premium feature.
This is not to say that safety is unimportant. On the contrary, the ethical implications of AI deployment are more critical than ever. As enterprises increasingly rely on AI for high-stakes decisions—in healthcare, finance, and criminal justice—the need for robust safety measures becomes paramount. But the market is telling us that safety must be integrated into the core product, not sold as an add-on. It must be a default, not a luxury.
My experience during the 2022 DeFi collapse reinforced this lesson. I witnessed the implosion of several lending protocols that I had previously advocated for, and the emotional exhaustion drove me to withdraw from public discourse for six months. During that time, I audited 12 failed smart contracts and identified a common thread: over-leveraged designs that ignored real-world utility for speculative yield. The protocols had focused on technical sophistication at the expense of practical value, and they paid the price. The same principle applies to AI models. A model that is technically brilliant but commercially unviable is a liability, not an asset.
The contrarian angle here is that Fable 5's low adoption rate may actually be a strategic advantage for Anthropic, not a weakness. By positioning Fable 5 as a premium product for specialized use cases, Anthropic can maintain its reputation as a frontier lab while generating revenue from Opus 5. This "decoy effect"—where the existence of a high-priced option makes the mid-priced option seem more attractive—is a classic pricing strategy. It is possible that Anthropic deliberately priced Fable 5 at a premium to make Opus 5's value proposition more compelling. If this is the case, the company is executing a sophisticated market strategy that prioritizes volume over margin.
But this strategy has a hidden cost. If enterprises perceive Fable 5 as overpriced and underutilized, they may question Anthropic's understanding of the market. They may also be reluctant to adopt future flagship models, fearing that they will be similarly overpriced. This could create a long-term trust deficit that is difficult to repair. Trust, after all, is the foundation of any lasting relationship, and it is built on consistent, predictable behavior. If Anthropic is seen as a company that does not understand the value of its own products, it will struggle to maintain customer loyalty.
The data from Ramp also raises questions about the broader AI infrastructure. If flagship models are underutilized, the demand for high-end GPUs such as NVIDIA's H100 may be lower than expected. This could have a ripple effect across the supply chain, affecting not only GPU manufacturers but also cloud providers and data center operators. The AI industry has been operating on the assumption that demand for compute will grow exponentially, driven by the need to train and run ever-larger models. But if enterprises are shifting toward mid-tier models, the growth in compute demand may be more moderate than anticipated. This is not necessarily a bad thing—it could lead to more efficient use of existing resources and a more sustainable growth trajectory. But it is a factor that investors and infrastructure providers must consider.
There is also a human dimension to this story. The enterprises that are choosing Opus 5 over Fable 5 are not making a purely rational economic decision. They are making a decision based on trust—trust that the mid-tier model will deliver sufficient value, trust that the cost savings can be reinvested elsewhere, and trust that the AI provider will continue to support and improve the product. This trust is built through experience, through successful deployments, and through transparent communication. Anthropic has an opportunity to build this trust by being more responsive to market feedback, by offering more flexible pricing, and by demonstrating a genuine commitment to customer success.
In my work as a decentralized protocol PM, I have seen firsthand the importance of aligning technical vision with market reality. The protocols that succeed are not necessarily the most technically sophisticated; they are the ones that solve real problems for real users. The same principle applies to AI. Fable 5 is a remarkable achievement, but it is not solving a problem that most enterprises are willing to pay for. Opus 5, by contrast, offers a compelling value proposition: near-frontier intelligence at a reasonable price. It is the model that most enterprises need, even if it is not the model that most excites them.
The path forward for Anthropic is clear. The company must embrace a multi-tiered strategy that recognizes the diverse needs of the enterprise market. It must continue to push the boundaries of what AI can achieve, but it must also be pragmatic about how it monetizes those achievements. Fable 5 should be positioned as a specialized tool for high-value, low-frequency tasks, while Opus 5 should be the workhorse that drives revenue growth. The company should also invest in building a more robust ecosystem, with better developer tools, more integrations, and more flexible pricing. And it should double down on its safety narrative, not as a premium feature, but as a core value that differentiates it from competitors.
This is not a story about failure. It is a story about adaptation. The AI industry is maturing, and the rules of the game are changing. The era of paying a premium for marginal intelligence gains is over. The era of value-based pricing has begun. Enterprises are no longer asking, "Which model is the most powerful?" They are asking, "Which model delivers the most value for my specific needs?" This is a more sophisticated question, and it demands a more sophisticated answer. Anthropic has the technology to provide that answer. The question is whether it has the strategic vision to do so.
As I reflect on my own journey—from the privacy-focused payment startup in Berlin to the DeFi collapse and the institutional bridge-building in Copenhagen—I am struck by a recurring theme. The technologies that endure are not the ones that are the most advanced; they are the ones that are the most trusted. Trust is not built on capability alone. It is built on reliability, on transparency, and on a genuine commitment to the well-being of users. Anthropic has an opportunity to build this trust by listening to the market, by adapting its strategy, and by demonstrating that it understands the value of its own products.
The future of AI is not a single model. It is a portfolio of models, each designed for a specific purpose, each priced according to the value it delivers. The enterprises that thrive will be those that learn to navigate this portfolio, selecting the right tool for the right task. The model providers that thrive will be those that help enterprises make these choices, offering guidance, flexibility, and support. This is the new paradigm, and it is one that demands a different kind of intelligence—not just technical intelligence, but market intelligence, strategic intelligence, and emotional intelligence.
Truth is not what is seen, but what is trusted. The data from Ramp is a visible truth, but the deeper truth is the trust that enterprises are placing in value over capability. This is a shift that will reshape the AI industry, and it is a shift that Anthropic must embrace if it is to remain a leader. The company has the technology, the talent, and the vision. What it needs now is the wisdom to align its strategy with the realities of the market. The path is clear. The question is whether Anthropic will have the courage to walk it.
In the end, this is not a story about a model that failed. It is a story about a market that is maturing, a market that is learning to distinguish between hype and value, a market that is demanding more from its technology providers. This is a healthy development, and it is one that will ultimately benefit everyone—providers, enterprises, and society as a whole. The AI revolution is not over. It is just getting started, and it is getting started on a more solid foundation. The era of blind performance-chasing is over. The era of thoughtful, value-based adoption has begun. And that is a future worth building.