Sonnet 5.5
- Source
- anthropic.com
- Author
- D2OQZG8l5BI1S06
- Date
Anthropic's Sonnet 5.5 keeps Sonnet 5 pricing at $2 input and $10 output per million tokens, runs more than 30% faster, and typically needs fewer tokens per task. Anthropic reports 70.6% on Terminal-Bench 4.0 and calls out large agentic coding gains. It is also the first Sonnet shipped with cyber safeguards, so some security requests may fall back, and Haiku 5.5 is due in the coming weeks.
Distribution was immediate. GitHub made it generally available in Copilot across the app, CLI and VS Code, reporting the same coding results as Sonnet 5 with fewer steps, tokens and tool calls. Cursor added it and says it matches Opus on many tasks. Cognition put it in Devin and reports 64.4% on FrontierCode 1.1 Main, up from 56.2% for Sonnet 5.
Simon Willison notes it is now the free-tier model, and flags one trap: a max-effort run consumed all 128,000 tokens, about $1.28, and returned nothing. The practical move is to rerun a representative slice of work you currently send to Opus on Sonnet 5.5 at a capped effort level and compare cost per completed task.
Context
Anthropic attributes Sonnet 5.5's lower task costs to using fewer , the units of text a model processes and generates. Its launch announcement keeps Sonnet 5's rates of $2 per million input tokens and $10 per million output tokens, while reporting costs up to 30% lower per task in its testing. The company positions the model for well-defined everyday work and says Opus 5.5 remains stronger on complex, open-ended assignments. The relevant comparison is therefore the cost of completing a particular job, including the selected reasoning effort.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
Checking sign-in…
Loading comments…