Deutsche Bank flags 65x AI cost gap between frontier and open models

Deutsche Bank estimates Anthropic's Claude Fable 5 costs about $3.25 per task, while DeepSeek V4-Pro runs at roughly 5 cents — a 65x gap that the bank says resembles "status good" pricing rather than a reflection of genuine performance superiority.

"Frontier models are a roaring brand-new supercar; open-weight models are a reliable second-hand family estate car," the analysts wrote in a June 20 report. For roughly 90 percent of routine enterprise tasks, the cheaper model delivers comparable results, they said.

Claude Fable 5 scores 60 on the Artificial Analysis intelligence index versus DeepSeek V4-Pro's 44. The cost of running AI at a fixed capability level has been falling by a factor of 9x to 900x per year, the report noted, while the lag between closed frontier models and the best open-weight alternative has compressed from roughly 12 months to about three months.

The finding threatens the pricing power of frontier labs such as Anthropic and OpenAI as they prepare for initial public offerings. If enterprises shift 80 percent of workloads to models that are 99 percent cheaper — as Coinbase Chief Executive Officer Brian Armstrong predicted last week — the revenue models underpinning multibillion-dollar AI valuations face structural pressure.

The Cost Chasm Widens

The gap is not merely a US-versus-China story. Meta's Muse Spark, Nvidia's Nemotron 3 Ultra and OpenAI's own gpt-oss-120b all sit in the low-cost tier alongside DeepSeek, the report said. The true dividing line runs between proprietary frontier models and open-weight alternatives, not between geographies.

Anthropic's pricing decisions illustrate the dynamic. Claude Fable 5 launched June 9 at $10 per million input tokens and $50 per million output tokens — double the price of Opus 4.8 and the most expensive major model on the market. The company then moved programmatic Claude usage onto metered credits billed at full API rates on June 15, delivering an effective price increase of 12x to 175x depending on the task, according to estimates cited in the report.

The shift from flat-rate subscriptions to per-token billing is exposing enterprise cost sensitivity. Uber burned through its entire 2026 budget for AI coding tools by April and now caps each employee at $1,500 per tool per month in token spend. ServiceNow blew through its full-year Anthropic budget in the first few months of 2026. Even Microsoft canceled most of its internal Claude Code licenses in May and moved engineers to GitHub Copilot.

The Revaluation Risk

Deutsche Bank draws a direct parallel to the "DeepSeek moment" in early 2025, when markets realized near-frontier AI capability could be built at far lower cost. That shock triggered a sharp selloff in AI stocks, though the market later recovered as overall demand continued rising.

The current reckoning may prove quieter but more lasting, the bank said. If proprietary AI models have been partly priced and traded as status goods — where high price is itself a feature — then a full market repricing of their cost-effectiveness could produce a second, deeper revaluation of AI equities.

Epoch AI research cited in the report provides independent corroboration: the US-China frontier AI capability gap averages about seven months, a spread that closely matches the gap between proprietary and open-weight models. The geopolitical and commercial dimensions of the AI divide are essentially the same chasm, the report concluded.

Anthropic closed a $65 billion funding round at a $965 billion valuation in late May, with run-rate revenue of $47 billion, up from $9 billion at the end of last year. The company has every incentive to protect the revenue on which its valuation depends, the report noted. But as viable substitutes proliferate — open-weight tokens cost 8 to 100 times less than frontier models — no enterprise needs to tie its operations to any single provider.

Companies that build model-agnostic routing layers, where each task is matched to the cheapest model that performs it adequately, will use more AI for less money and on their own terms. Enterprise procurement teams are already catching on: large companies spent $37 billion on generative AI in 2025, with more than half — $19 billion — going to the application layer rather than to model providers directly.

This article is for informational purposes only and does not constitute investment advice.