Claude Sonnet 5 API prices rise 50% in September — how to offset costs without cuts

Anthropic's introductory pricing for Claude Sonnet 5 ends on September 1, 2026, raising input costs from $2 to $3 per million tokens and output from $10 to $15, a 50% increase for all pay-as-you-go API users. Rather than downgrading models or removing features, developers can use two built-in API mechanisms — prompt caching and batch processing — to absorb much of the increase. Prompt caching allows stable content like system prompts or documentation to be stored and re-read at just 10% of the standard input price, meaning cache hits at $0.30 remain cheaper than the old full input rate of $2. To use it, developers mark a stable block with cache_control, after which subsequent calls within the TTL window are billed at the reduced cache-read rate. However, caching only activates above a 1,024-token minimum, requires a byte-identical prefix on every call, and fails silently, so monitoring the usage field is essential to confirm hits are actually occurring.