Frontier Models are Losing Corporate Budgets

Anthropic's most capable model, Fable 5, has plateaued at roughly 11% of corporate spending on the company's tools more than two months after launch, according to the Financial Times, citing Ramp's analysis of spending across 70,000 companies. Its cheaper sibling, Opus 5, has already overtaken it. A 656-point Hacker News thread reacting to the story turned into something more useful than stock commentary: hundreds of actual users explaining, in granular detail, why they've stopped reaching for Anthropic's best model even when they can afford it.

The pattern across the thread is consistent enough to treat as signal. It's not really about Fable being weak. It's about trust.

Pricing whiplash and safeguards are pushing users toward "good enough"

One top comment captured the mood: people want AI access "like electricity," dependable regardless of what's happening upstream, and Anthropic's repeated changes to usage limits and plan terms have instead made every session feel provisional. Multiple users described genuine account disruptions, one reporting their startup's account was banned without warning after a remote contractor logged in, taking weeks to restore. Others described Fable's safety classifiers flagging routine cybersecurity, biology, or code-review work, with one user saying it took a month to complete a single project review because of repeated interruptions. Commenters also flagged that Fable isn't available under a zero-data-retention arrangement in the way older models are, which multiple enterprise users named as an outright adoption blocker regardless of capability.

Against that backdrop, the practical response has been to stick with Opus 4.8, an older model users describe as reliably serviceable, or to route around Anthropic entirely toward OpenAI's Sol or Chinese open-weight models like Kimi K3 and GLM-5.3, which several commenters said now perform close enough to frontier quality for most daily work at a fraction of the cost. Accel partner Miles Clements offered the market-level version of the same point to the FT: "most people don't need to operate at the frontier."

What this means for how frontier labs compete next

The uncomfortable read for Anthropic, and for every lab chasing frontier-tier pricing, is that raw capability stopped being the deciding factor once "good enough" models got cheap and plentiful. Reliability, predictable pricing, data handling terms, and whether the tool trusts its own paying customers are now competing directly with benchmark scores for the same budget. Anthropic's overall business is still growing quickly, with annualized revenue reported at $65 billion in July, but the specific bet that customers would pay a premium to stay at the frontier is the part investors are now questioning ahead of a reported IPO near a $2 trillion valuation.

If your team is choosing AI tools based on benchmark headlines rather than what will actually hold up in daily use, this is the reminder to slow down. Our growth strategy work starts with matching the tool to the job, not the other way around, and our AI marketing services team can help you figure out which AI model actually earns its price for your workflow.