Anthropic halves Claude Sonnet 5.5's cache read price to $0.10
From October 7, 2026, prompt cache reads on Claude Sonnet 5.5 cost $0.10 per 1M tokens instead of $0.20. Input, output and cache write prices are unchanged.

On October 7, 2026, Anthropic cut the price of prompt cache reads on Claude Sonnet 5.5 from $0.20 to $0.10 per million tokens. Cache reads now cost 0.05x the input price instead of the usual 0.1x. Anthropic’s release notes say cache writes and all other prices stay the same.
Sonnet 5.5 itself launched on September 28, 2026, at $2 input and $10 output per million tokens, with a 1M-token context window and up to 128,000 output tokens.
What it costs now
| Per 1M tokens | Before Oct 7 | From Oct 7 |
|---|---|---|
| Input | $2.00 | $2.00 |
| Output | $10.00 | $10.00 |
| 5-minute cache write | $2.50 | $2.50 |
| 1-hour cache write | $4.00 | $4.00 |
| Cache read | $0.20 | $0.10 |
Batch requests are still 50% off input and output.
Who saves
A cache read is what you pay when a request reuses a prompt prefix you sent before: a long system prompt, tool definitions, a document, or the earlier turns of an agent’s conversation. Agents that loop through many tool calls resend the same prefix again and again, so most of their input is cache reads. In its Haiku 5.5 announcement, Anthropic says the cut lowers Sonnet 5.5’s cost on most agentic tasks by about 20%.
Apps that send short, one-off prompts without caching see no change.
How it compares
Claude Opus 5.5 already used the same 0.05x rate, which works out to $0.20 per million cached tokens on its $4 input price. Claude Sonnet 5 keeps the standard rate of $0.20. Claude Fable 5.1 charges 0.025x, or $0.25.
What to do
If you run Sonnet 5.5 without prompt caching, turn it on: the smallest prompt that can be cached is 512 tokens. Then check your usage page for the share of input billed as cache reads, which is the part of your bill this change reduces.
- Sonnet 5.5 cache read−50%Before$0.20Now$0.10
Sources
Read on . Prices are in US dollars. How we check facts