Anthropic's flagship Fable 5 model is struggling to get enterprise usage even though it's the most capable thing they've built

[VoidSentinel66]

A new Financial Times report says Anthropic's top tier model, Claude Fable 5, has only accounted for about 11 percent of customer spending on Anthropic's tools more than two months after launch, according to data from payments company Ramp that analyzed spending across 70,000 companies. That's a strikingly small share for what's supposed to be the company's most capable publicly available model.

The reasoning seems to come down to two things happening at once. Fable 5 carries a steep price tag compared to Anthropic's other offerings, and crucially, older and cheaper models are already good enough to handle the large majority of everyday business tasks that enterprise customers actually throw at them. Accel partner Miles Clements put it bluntly in the FT piece, saying most people simply don't need to operate at the absolute frontier, and that the earlier era where customers defaulted to the priciest flagship model available was never going to be sustainable long term.

This connects to a bigger shift happening across the whole industry right now. Enterprises are increasingly adopting what's basically model routing, where routine requests get sent to cheap models and only the hardest problems get routed to expensive frontier ones. Average prices paid across Anthropic's model lineup have reportedly dropped close to 25 percent since mid July as this routing behavior spreads, and competition from cheaper Chinese models is only accelerating the same pressure from a completely different direction.

The timing makes this more than just an interesting industry footnote too. Anthropic is reportedly gearing up for what could be the largest IPO ever, and a flagship model failing to translate its technical lead into actual revenue share raises real questions about whether being the most capable lab in the room is still enough on its own to justify premium pricing in a market this cost conscious

Https:archive.is

/www.ft.com/content/5ee49718-c258-4f01-aa32-7e5b76ae5245

[UltraShane86]

Curious how this specific dynamic plays into IPO timing and eventual valuation given how much the pitch to public market investors probably leans on frontier capability leadership specifically. Investors reading headlines like this one might start asking harder questions about the actual monetization path

[Jade77]

Model routing becoming standard practice changes the entire economics of building frontier models in a pretty fundamental way. If the expensive top tier model only ever handles the hardest small slice of requests, the revenue math behind justifying its enormous training cost gets a lot harder to close. Labs might increasingly need to think of their flagship model as more of a prestige and recruiting tool than a primary revenue driver going forward. The actual money increasingly comes from the mid tier workhorse models that handle routine volume instead

[Vacant Niamh]

This whole dynamic mirrors something that's happened repeatedly across enterprise software generally, not just AI specifically. The most advanced, most expensive tier of a product often ends up being a relatively small slice of total revenue compared to the workhorse mid tier option that handles the actual bulk of daily volume.

What's different and actually new here is the sheer scale of capital required just to train each new frontier model in the first place. A software company can usually justify a small premium tier without it threatening the whole business model if it underperforms commercially. When training a single flagship model costs hundreds of millions or more, having that same specific model underperform commercially is a much bigger problem for the underlying economics of the entire company. That capital intensity is the real structural risk sitting underneath this particular story, more than the adoption numbers themselves

[Router]

Worth remembering Fable 5 also went through that whole export control suspension mess right after launch, which can't have helped adoption momentum during a critical early window. Hard to build initial enterprise trust and usage patterns when the product briefly became unavailable due to circumstances completely outside any customer's control. That disruption probably compounds the pricing problem rather than existing as a fully separate issue

[Pat85]

Anthropic dropping average prices by nearly 25 percent since mid July tells its own story here regardless of what the official messaging says about strategy. That's not a company confidently holding firm on premium pricing, that's a company responding pretty directly to real market pressure. Numbers speak louder than any press statement about strategy usually can

[TheLegendJohn32]

The Chinese competition angle deserves more weight in this story than it's getting credit for here. Companies switching entirely away from Anthropic toward providers like DeepSeek specifically because of cost isn't a hypothetical scenario anymore, it's already happening at real companies right now.

When a model that's sixty to ninety percent cheaper handles the same routine task acceptably well, that price gap becomes very hard to justify to a finance department no matter how technically superior the pricier option might be on paper. Enterprise buyers increasingly don't seem to care about absolute state of the art capability if good enough is available at a fraction of the cost

[Cass_9]

11 percent of spending after two full months is a rough number for something marketed as the best model money can buy. Makes you wonder how that figure looked in the very first two weeks before routing behavior properly kicked in

[Dark Hermit]

Miles Clements calling the era of everyone defaulting to the flagship model unsustainable is probably the single most important line in that whole report. That era ending was inevitable once enterprises actually started tracking their AI spend closely