Anthropic launched Claude Sonnet 5.5 on September 28 for everyday coding and professional work. Its announcement claims faster generation and lower cost per task than Sonnet 5. That does not mean cheaper standard input and output tokens.
Price and capacity
Standard Claude API pricing remains $2 per million input tokens and $10 per million output tokens, with $0.20 cache reads. The model documentation lists a one-million-token context window and a 128K standard output limit. Batch beta output limits differ.
Check your integration before switching
The migration documentation identifies breaking changes around thinking, forced tool use, conversation-bound thinking blocks and older computer-use tooling. Apps use Medium effort by default; the API uses High. That distinction matters when comparing latency and task cost.
Anthropic's efficiency claims are provider-reported results. They do not establish a CompareLLM category ranking or guarantee savings on another workload.
Sources checked October 4: Anthropic announcement, model specifications, migration changes.