Speed runs require foresight, not just reaction. In the cryptocurrency world, I learned that lesson the hard way during the 2017 ICO frenzy—45 whitepapers in a month, and the ones that paid off weren't the flashiest, but the ones with sustainable tokenomics. Now, watching the AI model market, I see the same pattern. Anthropic's flagship Fable 5, launched with fanfare as the "strongest model," is floundering in enterprise adoption. According to Ramp transaction data, Fable 5 accounts for only 6% of Anthropic's token usage and a mere 11.4% of its spending. Meanwhile, the cheaper Opus 5—positioned as "near-Fable-5 intelligence at half the price"—has already surpassed it in just a few months. This isn't a technical failure; it's a market signal. The ledger does not lie, but it rewards patience. And the enterprise wallet is voting for value, not bragging rights.
Context: The Double-Edged Sword of Anthropic's Dual-Flag Strategy
Anthropic, backed by over $10 billion in funding and a reputation for AI safety, launched both Fable 5 and Opus 5 in mid-2025. The pitch was clear: Fable 5 for the cutting edge, Opus 5 for the practical. But the data from Ramp—a third-party payment processor used by thousands of enterprises—tells a different story. While Fable 5 was designed to command a 2x premium ($10/M input tokens, $50/M output vs. Opus 5's $5/$25), enterprises are overwhelmingly choosing Opus 5. On the other side, OpenAI's GPT-5.6 Sol already accounts for 25% of OpenAI's enterprise token usage, suggesting that the market's appetite for "best" models is real—but only when the price is right.
This isn't a repeat of the 2020 DeFi yield war, where hype drove unsustainable loops. This is a structural shift. Enterprises are no longer buying the "best" model; they're buying the model that delivers the best ROI for their specific workload. And that's a fundamental change for the AI industry.
Core: Why Fable 5 Failed the Enterprise Wallet Test
1. The Diminishing Returns Trap
Fable 5 is undeniably powerful. But the gap between it and Opus 5 is incremental, not generational. Anthropic itself admitted this by positioning Opus 5 as "near-Fable-5 intelligence." From my experience auditing DeFi protocols in 2020, I saw the same pattern: when a "v2" offers only marginal improvements over a "v1" but charges double, the market punishes it. The price elasticity is brutal. A 100% price increase for an estimated 10-15% capability gain? Enterprises are too sophisticated for that. The days of "build it and they will pay any price" are over.
2. The Token Usage vs. Spending Gap
Fable 5 uses 6% of Anthropic's tokens but accounts for 11.4% of spending. That means the few customers who use it are paying a premium, but the volume is too small to matter. Compare this to Opus 5, which likely accounts for the vast majority of the remaining 88.6% of spending. The enterprise is voting with its wallet, and the message is clear: "We want capabilities, not quadratically increasing costs."
3. The OpenAI Contrast
GPT-5.6 Sol's 25% token usage share among OpenAI's enterprise clients shows that flagship models can succeed. But the difference is likely pricing. OpenAI has historically been more aggressive in offering tiered pricing, and GPT-5.6 Sol may be perceived as delivering better value for its cost. Anthropic's premium pricing for Fable 5 may have simply been too high for the perceived gain.
4. The Industry Expert Consensus
Accel partner Miles Clements nailed it: "Most people don't need the cutting-edge model all the time." This is exactly what I saw in the crypto market during the 2022 NFT crash—the hype chains break when the numbers don't add up. Enterprises are now applying the same ROI logic to AI models that they apply to cloud computing or any other infrastructure: optimize for cost, not prestige.
Contrarian: The Unreported Angle—Fable 5's Low Adoption Is Healthy for the Market
Most analysts are calling this a failure for Anthropic. I disagree. The market is finally maturing. Here's what everyone is missing:
- Fable 5's low adoption proves that the AI model market is not a natural monopoly. Unlike the early days of blockchain where Ethereum's network effect made it nearly impossible to displace, AI models are increasingly commoditized. Enterprises can switch between Fable 5, Opus 5, GPT-5.6 Sol, or even open-source models like Llama 3.1 405B. This competition is healthy and will drive down costs for everyone.
- Anthropic's dual-flag strategy is actually working. Opus 5's rapid adoption shows that Anthropic has a viable growth engine. The company is not reliant on the flagship to generate revenue. This is a smarter play than OpenAI's all-in on GPT-5.6 Sol, which could become a cash cow but also a single point of failure.
- The "best model" narrative is losing its power. In 2017, I saw ICOs raise millions on whitepapers alone. Today, that doesn't work. Similarly, "best model" status without a clear use case is a marketing gimmick. Enterprises want models that solve specific problems—code generation, customer support, data analysis—at a predictable cost. Fable 5 may be overkill for 90% of those tasks.
- From the noise of 2017 to the signal of today, the market is shifting from "capability arms race" to "value layer competition." This is a bullish sign for the entire AI ecosystem. It means that startups and enterprises can build applications without worrying about being locked into a single expensive model. The commoditization of AI is accelerating.
Takeaway: What to Watch Next
The next three months will be telling. If Anthropic responds by cutting Fable 5's price or releasing a "Fable 5 Lite," it will validate the market's pricing power. If OpenAI raises GPT-5.6 Sol's price, it will test whether the flagship model has any pricing power left. And if open-source models like Llama 4 start capturing significant enterprise workloads, the closed-source premium will erode further.
Speed runs require foresight, not just reaction. The ledger does not lie, but it rewards patience. For now, the enterprise wallet is speaking clearly: value wins. The question is whether Anthropic and OpenAI will listen.