Claude Sonnet 5.5: cache reads cost half, from $0.20 to $0.10
Claude Sonnet 5.5 cut the price of reading cache in half: from $0.20 to $0.10 per million tokens. What changes for your agent and what does not.
Claude Sonnet 5.5 cut the price of reading cache in half: from $0.20 to $0.10 per million tokens. Anthropic noted it in its platform release notes on October 7, 2026. If your agent rereads the same context on every call, this is the line on your bill that just dropped.
What changed in Sonnet 5.5 cache pricing
Reading cache on Claude Sonnet 5.5 went from $0.20 to $0.10 per million tokens. That is 0.05 times the base input price, which is $2 per million. Put another way: reading from cache is 20 times cheaper than sending the same text again.
What you repeat on every call is what you pay most for, and that is where the discount lands.
What did not change
- Writing to cache is still $2.50 per million tokens (5-minute cache).
- Every other price is the same.
- It only applies if you already use prompt caching. If you do not, this cut does not touch you.
An example with numbers (WAiBOT calculation)
Picture an agent with a 50,000-token prompt that is reread 100 times. That is 5 million tokens read. Sent again at the input price ($2 per million) it would be $10. Read from cache at $0.10 per million, it is $0.50. Before the cut, that read cost $1. The example ignores the initial cache write and is only an illustration, not an Anthropic figure.
How to apply it in your business?
These ideas are WAiBOT suggestions, not official Anthropic recommendations:
- Check which part of your calls repeats: instructions, reference documents, long history.
- Cache that long prompt just once and read it from there.
- Instead of trimming your context to save money, keep it long and let the cache do the work.
- Compare your bill before and after to see how much it really dropped.
If you want to compare with another model in the family, read our breakdown of Claude Haiku 5.5: pricing and API fine print.
Frequently asked questions
How much does reading cache cost now on Claude Sonnet 5.5? $0.10 per million tokens, down from $0.20.
How much does writing to cache cost? Still $2.50 per million tokens, on the 5-minute cache.
How many times cheaper is a cache read? 20 times cheaper than sending the same text again at $2 per million.
Does it apply if I do not use prompt caching? No. It only applies if you already use prompt caching.
Conclusion
The cut is simple: reading cache on Sonnet 5.5 costs half, and for an agent that repeats context all day it can show up on the bill. Save this analysis for when your agent's bill arrives. Sources: Claude Platform release notes and Claude API pricing.