Skip to main content

Anthropic's Fable 5 Costs Triple to Run vs. Open Source Rivals

In the fast-paced world of AI, a new competitive dynamic is emerging. Wojciech Gryc recently took a close look at how Anthropic's Fable 5 stacks up against two new Chinese models: Kimi K3 from Moonshot, launched on July 16, and Qwen 3.8 from Alibaba, released on July 19. Both of these models deliver performance on par with Fable 5, and here's the kicker—they're both open-sourcing their weights.

Gryc points out that while training cutting-edge AI models eats up a lot of cash for researchers and computing power, the real ongoing expense is inference cost. Every time a user calls an API, it guzzles GPU resources and electricity. How companies handle these inference costs could make or break their business models.

He breaks AI companies into three buckets. First, there are those that rent data centers and pay the electricity bills—think Anthropic, Moonshot, and GLM 5.2. Second, companies like Meta and Alibaba build their own data centers. Third, there's SpaceX, which generates its own power and runs its own data centers. Gryc emphasizes that companies with solid infrastructure can still turn a profit by renting out servers, even if their models aren't the absolute best.

But here's where it gets interesting for Anthropic. Gryc notes that running Fable 5 for a single task costs nearly three times as much as OpenAI or open-source models. In a market where users are price-sensitive, can they stomach that kind of premium? He argues that benchmark tests are becoming saturated, and users perceive smaller differences between models than the cost gaps. That could spell trouble for Anthropic, potentially pushing customers toward cheaper alternatives.

Image

OpenAI, on the other hand, looks more competitive with its broad product lineup and data center investments. Gryc expects this wave of competition to keep rolling, with multiple labs—especially in China—potentially overtaking traditional US tech giants in the near future.

Key Points

  • Cost Disadvantage: Fable 5's inference cost is three times higher than OpenAI and open-source models, putting pressure on its market position.
  • Infrastructure Matters: Gryc categorizes AI firms by infrastructure ownership, showing that those with data centers or self-generated power have a cost edge.
  • Rising Competition: Chinese labs like Moonshot and Alibaba are rapidly catching up, offering similar performance at lower costs.