What changed
ClaudeDevs posted on 8 October 2026 at 20:04 UTC that Sonnet 5.5 cache reads had been cut in half. On the Claude Platform, the account says, cache reads are now $0.10 per million tokens. The same parenthesis lists input at $2 and output at $10 as current rates beside that price. The cut, as ClaudeDevs describes it, applies to cache reads. The post does not print the previous dollar amount. "ICYMI" and "now" frame this as a change already made.
Cache reads are tokens the API serves from prompt caching, a stored prefix reused on a later call. ClaudeDevs estimates Sonnet 5.5 "now runs ~20% cheaper on most agentic work," the account's phrase for jobs that resend that prefix across many steps. No cache-hit rate, token mix, or formula is in the post.
Who it applies to
People paying Claude Platform API rates for Sonnet 5.5 are the audience, especially when a large share of tokens are cache reads. Input and output in the same note stay at $2 and $10. ClaudeDevs does not describe those two rates as reduced.
The post's closing line is "API only, no change to Claude Code usage limits." Usage limits are the caps on a Claude Code subscription, separate from API token invoices. The Claude Code changelog mentions Sonnet 5.5 in a code-review note and lists Haiku 5.5, a different model, at $0.10/$0.50 per Mtok (million tokens). It does not state these Sonnet 5.5 rates.
How to check the rate
ClaudeDevs names no command, request header, model ID, or console control. The rates are presented as already in effect for Claude Platform API calls that use Sonnet 5.5. The post asks nothing of Claude Code users, because it says those usage limits are unchanged.
Check it on your own Sonnet 5.5 API usage: cache reads at $0.10 per million, with input at $2 and output at $10 in the same note.
What the post leaves open
The post states the halving and the $0.10 price, and omits the old dollar figure. Cache-write prices, batch rates, a long-context breakpoint, region, and a model ID are also absent. The 20 percent figure hangs on "most agentic work," with no workload behind it.
The Claude Code changelog does not list the three Sonnet 5.5 rates. They remain ClaudeDevs' announcement unless a Platform pricing page states the same numbers.
Source
Also from
Sources and editorial approach
This brief relies on the ClaudeDevs post published at 20:04 UTC on 8 October 2026 and on the Claude Code changelog as it stood later that evening. The changelog is the release-note page for the Claude Code app. It mentions Sonnet 5.5 in a code-review note and lists token prices for Haiku 5.5. It does not state Sonnet 5.5 cache-read, input, or output rates, a halving, or the agentic savings estimate. A separate Claude Platform pricing page was not covered.