The headline number is the one I trust least. Per-token prices didn’t move, $2 in and $10 out, so every promised saving rides on Sonnet 5.5 using fewer tokens for the same job. Whether that holds on my workloads is cheap to check: rerun last week’s agent tasks, the Astro build fixes and small scoped refactors, and compare the bill per task rather than per million tokens.
The part I’d watch is the new guardrail. Risky cybersecurity requests get automatically handed to a fallback model, and Anthropic admits legitimate ones will sometimes get caught. If you self-host and ask about firewall rules or exposed ports, some of those may land there. Worth knowing before you blame your prompt.
The story — Anthropic released Sonnet 5.5, claiming it runs over 30 percent faster than Sonnet 5 and cuts cost per task by up to 30 percent. By Anthropic’s figures it scores 70.6 percent on Terminal-Bench 4.0, ahead of Opus 5.5’s 66.4 percent. Pricing stays at $2 and $10 per million input and output tokens. It’s available on the Claude platform, Amazon Bedrock, Google Vertex AI and Microsoft Foundry (Source).