Claude Pricing Calculator (Opus, Sonnet, Haiku)
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
Claude Fable 5 Anthropic | $10.00 | $50.00 | 1000K |
Claude Mythos 5 (Glasswing) Anthropic | $10.00 | $50.00 | 1000K |
Claude Opus 4.8 Anthropic | $5.00 | $25.00 | 1000K |
Claude Opus 4.7 Anthropic | $15.00 | $75.00 | 1000K |
Claude Sonnet 4.6 Anthropic | $3.00 | $15.00 | 1000K |
Claude Haiku 4.5 Anthropic | $1.00 | $5.00 | 200K |
Claude Sonnet 4.5 Anthropic | $3.00 | $15.00 | 200K |
Which Claude model should you use?
Claude Sonnet 4.6 is the production default — best-in-class on long-context coding and agent workflows. Claude 4.5 Haiku ($0.80/1M input) is for high-volume chat, retrieval rewrites, and tool routing. Claude Opus 4.8 is reserved for the hardest single-turn problems where the extra $60/1M output is justified by measurably better answers.
Claude vs GPT-5 pricing
Claude Sonnet costs roughly 2.4x more than GPT-5.5 per token. The gap closes once you turn on prompt caching: a 50K-token system prompt cached across a 1M-call workload turns Claude into the cheaper option for many agent and long-context use cases.
How to lower your Claude bill
(1) Turn on prompt caching for every persistent system prompt — 90% off cached reads. (2) Use the Batches API for non-realtime traffic — another 50% off. (3) Route easier turns to Haiku and reserve Sonnet for turns that actually need the lift. Many production teams cut their Claude bill 60–80% with these three changes alone.