Official LLM API price list

A "0.3x rate" only means something against a baseline — this table is that baseline. Prices are taken from each vendor's own pricing page, at list price (excluding volume discounts), in USD per million tokens. Work out real cost in the rate calculator →

Anthropic

Model Tier Input $/M Output $/M Cached input Context Services Notes
Claude Opus 5.5
claude-opus-5-5
Flagship $4 $20 — 1000K 81 →
Claude Sonnet 5.5
claude-sonnet-5-5
Workhorse $2 $10 — 1000K 81 →
Claude Haiku 5.5 cheapest in family
claude-haiku-5-5
Lightweight $0.1 $0.5 — 1000K 81 →
Claude Fable 5
claude-fable-5
Reasoning $10 $50 — 1000K 81 →
Claude Opus 5
claude-opus-5
Flagship $5 $25 $0.5 1000K 81 → Batch 50% off; cache reads save up to 90%
Claude Sonnet 5
claude-sonnet-5
Workhorse $2 $10 $0.2 1000K 81 → Introductory price made permanent on 2026-08-10; cache read $0.20, cache write $2.50; batch 50% off
Claude Haiku 4.5
claude-haiku-4-5
Lightweight $1 $5 $0.1 200K 81 → The Claude lightweight tier: fast and cheap, suited to short outputs and high concurrency.
Claude Fable 5.1
claude-fable-5-1
Reasoning $10 $50 $0.25 1000K 81 → Cache read cut to $0.25 on 2026-09-01

Three traps in any API price list

  1. A cheap input price does not mean a cheap bill. Output tokens usually cost 4–6x input, so if you are generating long text or completing code, the output column is your bill.
  2. Cache pricing differs by an order of magnitude. For some vendors a cache hit costs 10% of list price, so if you use a fixed system prompt or repeated documents, work out the caching benefit first.
  3. Promotional prices expire. Models with a promotion window noted in the remarks column may double afterwards — budget at list price for anything long-term (the Notes column flags these).

Prices change as vendors adjust them; the verification date for each row is in its notes. Check the vendor's own pricing page again before making a production decision.