The Information Machine

@mcuban $0.50 per Mtok is a lot of money Mark. Are you considered cache hit on prefill? Or just output tokens?

SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-05-16

SemiAnalysis challenges Mark Cuban's AI token pricing figure of $0.50 per million tokens, asking whether that price reflects cache hits on prefill tokens or only output tokens — a distinction that significantly affects true cost.

Open original ↗

Appears in

Extraction

Topics: llm-pricinginference-costsprompt-cachingtoken-economics

Claims

  • $0.50 per million tokens is, by SemiAnalysis's assessment, a significant amount of money.
  • The cost characterization of LLM usage depends critically on whether cached prefill tokens are counted separately from output tokens.
  • Mark Cuban made a claim about AI token pricing that SemiAnalysis believes requires clarification on pricing methodology.

Key quotes

@mcuban $0.50 per Mtok is a lot of money Mark. Are you considered cache hit on prefill? Or just output tokens?