Anthropic's Opus 5 overtakes flagship Fable 5 as buyers trade down
Anthropic's pricing problem is now official: the company's cheapest new frontier model is beating its most expensive one in corporate wallets, less than a month after launch — awkward timing for a $2 trillion IPO.
Opus 5, released July 24 at a fraction of Fable 5's price, has overtaken the flagship in business spending within roughly a month, according to Financial Times reporting on Ramp data covering some 70,000 companies. Spending on Fable 5 has plateaued at around 11 percent of overall outlay on Anthropic tools more than two months after its June launch; Ramp's July breakdown put it at 6 percent of Anthropic tokens but 11.4 percent of attributed spending, with Opus 4.8 alone at 28 percent of spend. The FT frames this as breaking "a pattern of corporate users defaulting to the most powerful models" — Accel partner Miles Clements, whose firm has invested close to $1 billion in Anthropic, was blunter: most people don't need to operate at the frontier.
The material development since we covered the first warning shot — Anthropic's Fable 5 stalls as corporate AI spending hits a ceiling — is that the ceiling now has a direction: buyers are not just declining to upgrade to Fable 5, they are actively routing budget below it. Independent data points agree. Vercel's gateway sample puts Fable 5 at 13.2 percent of model spending, and Simon Willison's read of Ramp's July numbers shows the flagship trailing even mid-generation models like Sonnet 4.6. Analysts told the FT that price plus the fact that older models handle most business demands explains the shift — and the timing compounds it, since Anthropic's bankers have floated an IPO of $100 billion-plus at a valuation of at least $2 trillion, which would be the largest on record. A listing pitched on frontier pricing power now has public evidence that its priciest product is its slowest seller. Anthropic declined to comment to the FT.
The take: this looks less like a demand problem than a routing one — when switching costs are near zero, "good enough" wins the bulk of tokens, and the flagship becomes a specialist tool for long-running autonomous work rather than the default meter running for everything. That reframes what frontier-lab revenue is worth: not a premium on every token, but a toll on the hardest slice of them.
Open weights are taking the displaced traffic: open models jumped from 28 percent to 62 percent of tokens on Vercel's gateway in two months, per investor Gavin Baker citing founder Guillermo Rauch. Baker notes OpenAI and Anthropic's combined usage still accelerated in July — the market grew fast enough that both things are true at once. It aligns with the routing story above: cheap Chinese and community models like GLM 5.2 (roughly a ninth of Fable 5's cost) absorb the routine coding work that flagships used to get by default.
SoftBank is set to sell a record 1 trillion yen (about $6.8 billion) of bonds aimed at funding its AI push, Japanese media report. The issuance targets retail investors and underlines how far Masayoshi Son is going to keep bankrolling OpenAI-related bets — borrowing at scale rather than spending cash on hand. Watch whether other AI-heavy conglomerates follow with similar retail-funded vehicles.
What to watch: whether Anthropic adjusts Fable 5 pricing or bundles it into enterprise plans before the IPO roadshow — and whether the next Ramp snapshot shows Opus 5's share still climbing or plateauing the same way.
Has your team downgraded from a flagship model to a cheaper one without noticing a quality drop? Tell us in the comments.
Sources: Financial Times via Techmeme · PYMNTS · Implicator.ai · Simon Willison's Weblog · Drew Breunig · 共同社
