Study: Anthropic models may be cheaper than some open-source Chinese models

rohanpaul_ai · x · 2026-08-15

A new study by AlphaSense suggests Anthropic models may be cheaper to use than some large open-source Chinese models. While Kimi is cheaper per token, it consumes many more tokens during context assembly, raising the total cost per question. The real cost metric is tokens-to-completion × token price. Efficient retrieval, stopping, and reasoning can lower the total inference bill. AlphaSense achieved a 3× cost reduction by optimizing retrieval, implying that controlling retrieval and routing decides economic viability.

Related event: US AI models cost less to use than Chinese rivals as context becomes the bottleneck(3 posts)→

Original post →

More from Infra

Infra channel →