The $10 per million input tokens question: why are enterprises abandoning the flagship?
The numbers landed like a cold splash of water on the AI industry's face. Fable 5, Anthropic's most powerful model ever released, accounts for just 11.4% of the company's enterprise spending through Ramp's payment platform. Six percent of token usage. That's it. The model that was supposed to define the frontier of artificial intelligence is being treated like an expensive experiment that enterprises peek at, then politely decline.
Meanwhile, Opus 5 — positioned as "near-Fable 5 intelligence at half the price" — is eating the lunch that was supposed to belong to its more sophisticated sibling. In less time on the market, it has captured nearly 89% of Anthropic's enterprise spend.
This is not a story about technical failure. This is a story about pricing psychology, value perception, and the moment when the AI industry realized that "best" and "most profitable" are two very different things.
I've spent the last nine years watching blockchain projects die from exactly this disease: building something technically magnificent while forgetting that adoption requires perceived value, not just actual capability. The Fable 5 situation is Déjà vu with different architecture.
The Data That Exposed the Emperor's New Pricing
Let's start with what we actually know. Ramp, the corporate spend management platform, released data showing token usage and spending patterns across Anthropic's model lineup. The numbers are stark:
- Fable 5: 6% of token usage, 11.4% of spending
- Opus 5: The overwhelming majority of both metrics, despite launching later
The pricing structure tells the rest of the story. Fable 5 commands $10 per million input tokens and $50 per million output tokens — exactly double Opus 5's rates. In a rational market, that premium would require a corresponding leap in perceived capability. But the data suggests enterprises looked at the 2x price tag, evaluated the marginal intelligence gain, and collectively said: "No thanks."
Here's the uncomfortable truth that model providers don't want to acknowledge: we have hit the diminishing returns curve, and it's steep.
Going from an 85 to a 90 on a benchmark test might require doubling the compute. Going from 90 to 92 might require quadrupling it. But the enterprise customer doesn't care about your training FLOPs — they care about whether the output quality justifies the API bill. And for most real-world workloads, the gap between Fable 5 and Opus 5 is simply not worth $40 per million output tokens.
This mirrors something I observed in the DeFi space repeatedly: the best technical solution doesn't win if the fee structure doesn't match perceived value. I've watched lending protocols with superior architecture lose to clunkier competitors because they insisted on premium fees for marginal improvements.
The Psychology of Enterprise AI Procurement
Accel partner Miles Clements nailed it when he observed that "most people don't need to constantly use frontier models." This sounds obvious, but its implications are profound.
Enterprise AI procurement has shifted from a technology-driven decision to a finance-driven one. The question is no longer "which model is smartest?" but rather "which model delivers adequate intelligence at the lowest total cost of ownership?"
This is a fundamental shift in how the AI market operates. During the early phase of the LLM boom, enterprises chased capability because the gap between models was massive. GPT-3 to GPT-4 felt like a generational leap. But Fable 5 to Opus 5? That's a half-step. An increment. An optimization rather than a revolution.
The data confirms this perception gap. Fable 5's token usage being only 6% while commanding 11.4% of spending suggests it's being used for specific high-value tasks — complex reasoning, long-document analysis, critical decision support — while the bulk of enterprise workloads run on the cheaper model.
This is actually rational behavior. Enterprises are essentially saying: "We'll pay premium prices for premium tasks, but we won't pay premium prices for everything."
But here's the kicker: the premium tasks apparently aren't numerous enough to justify Fable 5's existence as a mass-market product.
The Competitive Landscape: OpenAI Is Winning the Flagship Game
The contrast with OpenAI's GPT-5.6 Sol is instructive. Sol accounts for 25% of OpenAI's enterprise token usage — more than four times Fable 5's share relative to Anthropic's total.
This isn't because Sol is dramatically smarter than Fable 5. It's because OpenAI either priced it more sensibly or communicated its value proposition more effectively. Or both.
In the blockchain world, I've seen this pattern repeatedly: the project with better technology loses to the project with better market positioning. Technical superiority is necessary but never sufficient. The market rewards perceived value, not actual capability.
Anthropic's decision to position Opus 5 as "near-Fable 5 intelligence at half the price" was a strategic acknowledgment that they anticipated this problem. It's essentially saying: "We know our flagship is too expensive for most use cases, so here's the sensible option."
But this creates an awkward tension. If Opus 5 is "near-Fable 5," then why does Fable 5 exist? What's the actual use case that justifies the premium?
The answer appears to be: very few, according to the spending data.
The Infrastructure Trap: When Compute Becomes a Liability
Here's something the public data doesn't show, but my experience in high-cost technical infrastructure tells me is happening behind the scenes: Anthropic is likely bleeding money on Fable 5 inference.
Training a frontier model requires massive compute. Deploying it at scale requires even more. If Fable 5's inference costs are significantly higher than Opus 5's — which the 2x pricing suggests — then low utilization rates mean Anthropic is eating those costs without corresponding revenue.
This is the classic infrastructure trap. You build for scale, but if adoption doesn't materialize, your fixed costs become anchors. I've seen this in blockchain infrastructure projects that built for millions of transactions per second and then watched their token usage crawl at fractions of a percent of capacity.
The "smart" move for Anthropic would be to repurpose Fable 5's compute allocation toward Opus 5, which is clearly driving revenue. But that creates a strategic problem: Fable 5 is their "frontier" narrative. It's the model that justifies their technical leadership claims. Killing it or downgrading it would be an admission that the frontier isn't where the money is.
The Industry-Wide Implication: Value Stratification Is Here
This Fable 5 situation isn't just an Anthropic problem. It's a signal that the entire AI industry is entering a new phase: value stratification.
We're moving from a "capability arms race" to a "value segmentation" market. Model providers will need to offer multiple tiers that map to different use cases and budget constraints. The "one flagship model to rule them all" approach is dying.
This mirrors what happened in the blockchain space around 2020. For years, projects competed on raw technical specs — TPS, block size, consensus efficiency. Then the market matured, and we saw value stratification: DeFi protocols for yield farmers, payment rails for remittances, storage networks for enterprises. The projects that won weren't necessarily the most technically advanced — they were the ones that mapped most cleanly to specific user needs at specific price points.
For AI, this means:
- Model routing will become a standard practice — enterprises will use multiple models, selecting based on task complexity and cost sensitivity
- The middle layer will grow — tools that optimize model selection and cost management will become essential infrastructure
- Open source models gain credibility — when enterprises realize they don't need frontier capabilities for most tasks, open source options look increasingly attractive
The Contrarian Angle: Fable 5's Low Adoption Might Be a Feature, Not a Bug
Let me play devil's advocate for a moment, because I think there's a reasonable argument that Fable 5's low adoption rate isn't a failure — it's a deliberate strategy.
Consider this: Anthropic knows that the frontier model market is small. They know that only a tiny fraction of enterprise workloads require the absolute best reasoning capabilities. But they also know that having the "best model" is essential for their brand narrative and for attracting top-tier clients who want bragging rights.
By pricing Fable 5 at 2x Opus 5, they're essentially creating a prestige product that serves a marketing function while the real revenue engine runs on the mid-tier model. This is similar to how luxury car manufacturers have halo models that generate few sales but elevate the entire brand perception.
The risk, of course, is that Fable 5's low adoption rate becomes a self-fulfilling prophecy. If enterprises perceive it as "not worth the premium," the narrative shifts from "Fable 5 is the frontier" to "Fable 5 is overpriced." That's a dangerous transition.
The Regulatory and Security Dimension
There's another angle that hasn't received enough attention: the security implications of enterprises defaulting to mid-tier models.
Anthropic has built its brand around AI safety. Fable 5 presumably includes the most advanced alignment techniques, red-teaming, and safety guardrails. If enterprises are rejecting it on price grounds, they're also rejecting the safest model available.
This creates a troubling dynamic where the most secure AI systems are underutilized in favor of cheaper, potentially less aligned alternatives. It's the AI equivalent of choosing a cheaper car without airbags because the safer option costs twice as much.
I'm not suggesting this is a crisis — Opus 5 is still significantly safer than most alternatives. But it does highlight a tension between safety as a value proposition and safety as a cost center. Enterprises talk about wanting safe AI, but when the bill arrives, they optimize for cost.
What Happens Next: Three Scenarios
Scenario 1: Anthropic adjusts Fable 5's pricing The most likely response. Either a direct price cut or the introduction of a "Fable 5 Lite" tier that captures some of the premium capabilities at a more accessible price point. This would acknowledge the market's judgment while preserving the flagship narrative.
Scenario 2: Anthropic doubles down on Opus 5 Shift the entire enterprise strategy around the mid-tier model, treating Fable 5 as a research showcase rather than a commercial product. This would be a pragmatic admission that the frontier model market is tiny.
Scenario 3: The market splits further Other providers enter the mid-tier space aggressively, forcing Anthropic to compete on multiple fronts simultaneously. This would accelerate the value stratification trend and compress margins across the industry.
My bet is on a combination of scenarios 1 and 2. Anthropic will quietly reduce Fable 5's prominence in enterprise sales conversations while positioning Opus 5 as the default recommendation for most workloads.
The Takeaway: Trust the Process, But Verify the Code
The Fable 5 story is a reminder that in technology markets, the best product doesn't always win — the best-priced product does. Enterprises have spoken with their wallets, and they've said that frontier intelligence isn't worth the premium for most workloads.
This doesn't mean frontier models are irrelevant. It means their market is smaller than the hype suggests. The future belongs to providers who can offer layered solutions that match capability to cost, not to those who simply claim the crown of "most powerful."
Trust the process of market discovery, but verify the code — and the pricing — before you commit your enterprise budget.
The question now is whether Anthropic's leadership has the humility to listen to what the data is telling them, or whether they'll double down on a flagship that enterprises have already judged. The next quarter's numbers will tell us everything.