Anthropic's Fable 5 Falls Short on Enterprise Adoption

Two months ago, Claude Fable 5 launched with the kind of fanfare usually reserved for a flagship phone. Anthropic called it the most capable model on the market. The U.S. Department of Commerce briefly suspended access over export-control concerns, which only added to the buzz when it returned three weeks later. Then the spending data came in, and it told a much quieter story.

According to Ramp's August AI Index, which tracked spending across 70,000 businesses, Fable 5 accounts for just 11.4% of what companies spend on Anthropic's models and only 6% of the tokens they actually use.

Compare that with OpenAI's flagship, GPT-5.6 Sol, which captures 25% of tokens and 23% of spend on OpenAI's platform. Fable 5 pulled in only about three-quarters of the revenue GPT-5.6 Sol generated in July, despite costing roughly twice as much per token.

With just a few months to go before Anthropic IPO, Ramp's lead economist, Ara Kharazian, didn't mince words on social media, saying Fable 5 "disappointed both in adoption and real-world application" given its price and data-handling requirements. He credited OpenAI's newer model with being genuinely strong and increasingly the default pick among developers.

Fable 5 runs at $10 per million input tokens and $50 per million output tokens through the API — twice what Claude Opus 4.8 costs and twice the price of GPT-5.6 Sol. Ramp's researchers read the plateau as evidence that businesses have hit a ceiling on what they'll pay for incremental gains in benchmark performance. Better test scores, in other words, no longer sell themselves.

Cost isn't the only sticking point. Fable 5 ships with mandatory 30-day data retention tied to the safety classifiers that route higher-risk cybersecurity and biology-related requests. For companies handling sensitive client data, that retention window is a compliance headache many would rather sidestep, and it keeps coming up as a reason teams stay with older, less restricted models instead.

Perhaps the clearest sign of trouble: Claude Opus 5, a cheaper model Anthropic launched in late July, has already overtaken Fable 5 in enterprise spending. Businesses aren't necessarily fleeing Anthropic for rivals; plenty are simply downgrading within Anthropic's own lineup, picking whichever model gets the job done for less.

None of this means Anthropic is collapsing. The company still led OpenAI in overall U.S. business AI spending through much of the summer, reaching a nearly 44% share in July versus OpenAI's roughly 40%.

But the most recent readings suggest that lead is narrowing again as GPT-5.6 Sol gains ground with developers, which is good news for OpenAI ahead of its upcoming IPO. Analysts caution against pinning too much on any single model's rollout: enterprise buying decisions turn on procurement policy, existing contracts, and integration effort as much as on raw capability.

Ramp's dataset also skews toward mid-market companies paying through its corporate card, leaving out large enterprises that route their spend through providers like American Express.

The takeaway isn't that frontier AI has stopped mattering to businesses. It's that "frontier" no longer guarantees automatic adoption. Companies are now treating each new model release as a purchasing decision to be weighed against its price and restrictions rather than as a default upgrade.

For an industry whose valuations lean on the assumption of ever-rising willingness to pay for ever-more-powerful models, that's a shift worth watching closely.