Claude Sonnet 5 Wants to Make Opus Optional

Anthropic shipped Claude Sonnet 5 on June 30, and the pitch is blunt: the mid-tier model now does most of what you used to pay flagship prices for. It plans, it executes across browsers and terminals, it writes and debugs code, and it checks its own work without you asking. Early partners kept repeating the same line, that it gets more done with less and finishes tasks where the older models would quietly give up halfway.

The number that matters is price. Through August 31 it runs at $2 per million input tokens and $10 per million output, settling to $3 and $15 after. That is a fraction of Opus money, and Anthropic is openly saying Sonnet 5 now approaches Opus 4.8 on reasoning, tool use and coding. For anyone running agents at scale, where the bill is the whole game, that gap closing is the actual headline. The flagship stops being the default and becomes the thing you reach for only when you truly need it.

There is a quieter and more interesting decision buried in the safety notes. Sonnet 5 was deliberately not trained for cybersecurity and shows sharply reduced exploit-development ability compared to the Opus line. Read that next to the whole government-gating story of the last month, where the dangerous offensive-cyber capability is exactly what gets locked behind approval. Anthropic is effectively building its cheap workhorse to be safe by subtraction, and keeping the loaded gun in Opus and Mythos. The cheap model you give everyone is also the model you made boring on purpose.

This is the second major Anthropic model in three weeks after Fable 5, and the cadence itself is the story. Sonnet 5 is live now across Free, Pro, Max, Team, Enterprise, Claude Code and the platform. Details at https://www.anthropic.com/news/claude-sonnet-5