Token Pricing
The per-million-token rate a provider charges for input and output. Expressed as dollars per million tokens. Claude Opus output is $75 per million tokens. Haiku is $5. The gap is fifteen times, which is what pricing tiers exist to navigate.
Providers publish token prices in dollars per million tokens. The numbers look small, but they add up fast. Claude Opus 4.7 at $15 input and $75 output means a single agent turn with 20,000 in and 5,000 out runs about $0.68. At 1,000 turns a day, that is $680 per day per feature, or roughly $20,000 per month. Cheaper tiers exist for a reason.
The Simple Version
Each model family has tiers. Opus is the smartest and most expensive. Sonnet is balanced. Haiku is fast and cheap. The same prompt costs dramatically different amounts depending on which tier it routes to. The operator’s job is matching the task to the cheapest tier that can still do it well.
Why It Matters
Using Opus for tasks Haiku can handle is how startups burn their runway. Most production workloads should be Haiku or Sonnet. Opus should be reserved for work that genuinely needs the extra capability: complex reasoning chains, high-stakes decisions, long autonomous runs. Pricing tiers are a discipline, not a menu.
How It’s Used on This Site
TWO’s digest writer agent runs on Opus because editorial quality benefits from it. The research agent runs on Sonnet because volume matters more than perfection. The tokenizer-level text cleanup runs on Haiku. Three tiers, three costs, one pipeline. Match the task to the tier.