Anthropic built Fable 5 to be its most capable model yet. Enterprises largely don’t care. More than two months after release, the model has captured only 11% of corporate spending on Anthropic’s products — a number that has barely moved since the first month after launch. That’s not a slow start. That’s a rejection, and it reflects something broader happening across the AI market right now.
The Financial Times reported the findings, citing data from Ramp, a payment services provider that tracks AI spending across 70,000 U.S. companies. The picture the data paints is clear: enterprises are done reflexively buying the most powerful model available. They want the best performance for what they’re actually paying per token.
Fable 5 is losing to Anthropic’s own cheaper model
The more embarrassing story isn’t that OpenAI is outcompeting Fable 5. It’s that Anthropic’s own Opus 5 is. Released in late July, Opus 5 is a smaller, cheaper model that has already surpassed Fable 5 in enterprise spending. Per Ramp’s data, it costs roughly half as much as Fable 5 and sometimes scores better on programming and knowledge-work benchmarks. So enterprises are getting more for less, without switching vendors.
Fable 5’s pricing makes the math brutal. Its per-token price is double that of Anthropic’s Opus 4.8 and ten times that of Haiku 4.5. For most enterprise workloads, that premium is hard to justify when a smaller model gets the job done. The old assumption that companies would always reach for the most powerful tool is clearly breaking down.
OpenAI’s GPT-5.6 is also pulling ahead
OpenAI has its own answer to Fable 5’s struggles: GPT-5.6, released last month at a significantly lower price point. According to Ramp tracking data, GPT-5.6 captured 25% of enterprise token usage and 23% of spending during the same period Fable 5 sat at 11%. Fable 5’s total revenue contribution was only about 75% of GPT-5.6’s. Given that Fable 5 costs twice as much per token, that gap in actual usage is even wider than the revenue numbers suggest.
Ramp’s chief economist Ara Karazian put it plainly: extrapolating previous trends would have suggested Anthropic would dominate the market, but with GPT-5.6 performing well and Fable 5 underperforming, the result has gone the other way. He also said predicting Anthropic’s trajectory over the next few months is “nearly impossible.”
Open-source models are piling on
The pressure doesn’t stop with proprietary competitors. Open-source models have surged from 11% of token usage in April to 62% by August, according to Vercel AI Gateway data. They still account for less than 4% of enterprise AI spending, so they’re not yet a revenue threat. But the trajectory matters. As open-source quality closes the gap with frontier models, the justification for paying premium prices gets harder to make.
The implication for companies like Anthropic, which have spent billions training increasingly large models, is uncomfortable. Miles Clements, a partner at Accel, which has invested approximately $1 billion in Anthropic, said most enterprises don’t need AI at the frontier level. He described the period when customers defaulted to only the most advanced models as “not a sustainable period.” He also noted that top-tier models may increasingly function as technical demonstrations rather than drivers of real usage volume.
Anthropic’s business is growing, but the IPO timing raises questions
None of this means Anthropic is in trouble. Its annualized revenue reached $65 billion in July, up from $47 billion in May, and the company posted its first adjusted operating profit in Q2. It now has 6,000 customers spending more than $100,000 annually. The company is also preparing for what could be a historic IPO, with investors valuing it at more than $2 trillion and a potential listing as early as next month.
But the Fable 5 numbers land at an awkward moment. A $2 trillion valuation is hard to sustain if your flagship product is being skipped in favor of cheaper alternatives. The key questions heading into that IPO are straightforward:
- Can Anthropic’s lower-cost models drive enough volume to offset weak Fable 5 adoption?
- Will the open-source surge eat further into the use cases that currently justify enterprise spending?
- Does the frontier model strategy remain viable if enterprises consistently choose value over peak performance?
The AI market is not moving toward a single dominant model. It’s fragmenting by price tier and use case. That’s good for buyers. For vendors betting on premium positioning, it’s a harder road than it looked a year ago.
