SAN FRANCISCO, Oct 8 — Anthropic has announced a pricing update for prompt cache reads on Claude Sonnet 5.5. The cache-hit read cost is halved from $0.20 to $0.10 per million tokens. Base input, output and cache write pricing remain unchanged.
Updated pricing for Sonnet 5.5:
• Base input: $2 per million tokens
• Base output: $10 per million tokens
• 5-minute cache write: $2.50 per million tokens
• 1-hour cache write: $4 per million tokens
• Cache read (hit): $0.10 per million tokens (50% reduction)
Prompt caching is designed for long-context workloads with repeated system prompts and large knowledge bases. When developers send identical long prefixes repeatedly, the system retrieves cached content instead of reprocessing the full prompt, cutting costs for recurring long-text requests. The price cut benefits enterprise users, RAG pipelines, document analysis and AI agents that rely heavily on long context.
Anthropic notes Sonnet 5.5 delivers over 30% faster inference compared to Sonnet 5. Combined with the cheaper cache reads, the update reduces total cost of ownership for enterprise long-context AI deployments. Industry observers see this pricing move as a response to competition from OpenAI and Google, aimed at capturing enterprise demand for RAG and long prompt applications.