Chinese models are now far cheaper per token

blaizedsouza · x · 2026-07-20

A repost argues that Chinese models are now dramatically cheaper than Anthropic on a per-million-token basis, citing a CNBC segment with Chamath.

The comparison given is roughly:

The main point is that inference costs are collapsing extremely fast, and premium pricing may only hold temporarily while companies trade on trust, safety, and enterprise relationships. The post stresses that these cheaper models are already shipping, not just benchmarks or demos.

Original post →

More from Models

Models channel →