Definition

LLM pricing is the effective cost of using large language models, driven by list price per million tokens and tokenizer behavior (how many tokens the same text produces). Cross-vendor comparisons require normalizing token counts, not only sticker rates.

Key Points

  • Tokenizer changes can inflate bills ~1.0–1.35× (Anthropic) even when list prices stay flat
  • Intro/promotional rates may mask post-promo effective cost increases
  • Playcode (2026-07): Claude new tokenizer up to 1.73× vs GPT-5.x on TypeScript
  • Production teams should measure token delta on own corpora before migrations

Sources