Model price comparison

In/Out are dollars per million tokens; Ctx is the context window. Worked example: at $3/M input, 1M input tokens = $3.00. Output rates are typically several times input rates, so a chatty model on a verbose workload can cost more than a pricier model that answers tersely. Rates in this table are maintained and revalidated, but providers change pricing — verify before committing a budget.

Why are input and output priced separately?

Providers bill the two lanes at different rates, and workloads skew differently: agentic coding reads far more than it writes, while generation workloads are output-heavy. Comparing on input price alone routinely picks the wrong model.

What about cache pricing?

Not shown here. Providers that discount cache reads (Anthropic prices them at roughly 10% of the input rate per their pricing page) can change the effective cost dramatically for stable-prefix workloads — use the token calculator, which includes cache lanes where we have rates.

Which model is cheapest for me?

Depends on your token mix. Take your real input and output counts (from a usage dashboard or local logs) and run them through the token calculator against each candidate — the ranking often flips between read-heavy and write-heavy workloads.

What are the limitations?

The table shows list rates only: no batch-API discounts, volume tiers, long-context surcharges, or subscription plans. Treat it as a comparison aid, and verify against your provider invoice.

When should I use this vs the full audit?

Use this to pick a model. The full stack audit addresses the other half of the bill — how many tokens your setup sends per request — which usually moves cost more than switching models does.