A scathing analysis published this week on leoveanu.com is making waves among developers who have noticed what many AI insiders have suspected for months: the math no longer works for American frontier models. The piece, titled around Qwen 3.7 Max as its centerpiece, dismantles the narrative that US labs deserve their pricing multipliers by introducing what the author calls the 'OnlyFans economy'—a framework explaining how premium AI customers are essentially paying cartel prices while receiving increasingly commodity results.

The $500 Million Problem

The author's most striking claim involves corporate AI adoption gone wrong. According to the analysis, some organizations have burned through '$500 million on Claude AI in one month' due to improper usage limits being configured, or consumed their entire 2026 AI budget within four months after laying off staff under the assumption that automation would pick up the slack. 'A life-changing amount of money was wasted on tokens that did not produce anything of value,' the author writes, adding these as early loser case studies for companies to study before repeating the same mistakes.

Qwen 3.7 Max and Extended Thinking

The Chinese model family has become a 'legacy for open-source,' according to the analysis, with Qwen 3.7 Max offering what US competitors have fragmented across multiple tiers: native extended-thinking that users can toggle on and off without navigating confusing pricing tiers like xhigh, max, ultracode, or medium variants. The author emphasizes that Chinese models aren't designed for one-shot cleverness demonstrations but rather 'the kind where you leave a model for hours and come back to find it actually finished the job.' This practical orientation stands in stark contrast to Anthropic's Opus 4.7 rate limits that sparked developer backlash upon release.

Developer Adoption Signals

Beyond personal benchmarks, the author points to OpenRouter rankings as 'no better signal than one coming from a large enough sample of what people actually use—those developers are voting with their wallets too.' The Artificial Analysis Intelligence index provides additional validation, though the author notes its scores are 'highly non-linear' and that recent Chinese models have narrowed gaps previously justifying geographic price premiums. When Anthropic suggests the AI trajectory 'may' turn out to be an S-curve while simultaneously pushing code as gospel, observant developers notice the plateau—the bend in the curve where US frontier models stopped earning their multipliers.

The Parasocial Tax

The piece draws a direct line between premium valuations and what the author calls parasocial economics. 'Nobody is simping for a cartel,' the analysis states, but the combination of premium-paying customers and devoted users maintaining brand loyalty regardless of performance creates inflated IPO expectations that will ultimately damage average American retirees whose index funds hold these positions. The author's conclusion frames this as a reckoning where 'when gravity finally asserts itself,' the consequences will extend far beyond tech Twitter discourse.

Key Takeaways

  • Chinese models like Qwen 3.7 Max now compete directly with US frontier offerings at a fraction of the cost
  • Corporate AI budgets are being decimated by poor implementation and missing guardrails on usage limits
  • OpenRouter rankings show developers increasingly choosing non-premium alternatives
  • The pricing premium for 'made in USA' no longer correlates with measurable intelligence gains

The Bottom Line

The OnlyFans economy metaphor cuts to the heart of it: premium AI customers have been paying cartel prices while getting rate-limited and receiving responses that increasingly come from models indistinguishable in capability from open-source alternatives. Developers who haven't tested Qwen 3.7 Max or equivalent Chinese models owe it to their organizations to run the comparison before signing another enterprise contract.