Anthropic released Claude Sonnet 5 on June 30. The launch line is that it is the most agentic Sonnet ever built, with real gains on browser control, terminal use, and multi-step tool loops that used to require larger and more expensive models. The interesting number is the price. Introductory API is $2 per million input tokens and $10 per million output tokens through August 31. After that it steps up to $3 in and $15 out, still below Opus by a wide margin. Sonnet 5 becomes the default model on Free and Pro, ships across Max, Team, and Enterprise, and drops straight into Claude Code.
Read this as the mirror image of the story that landed last week. GitHub Copilot spent June migrating every paid tier to metered billing, and the first month-end invoices sent developer chats into shock as agentic sessions blew past the old flat rate by ten or twenty times. Anthropic looked at that reaction, at Cursor’s usage-based pricing, at OpenAI’s gated enterprise Codex tier, and picked a different lane. Cheaper agent model, more of it, defaulted on for anyone with a Claude account.
The commercial subtext is louder than usual because Anthropic is running toward an IPO with what several outlets are now calling the busiest revenue quarter in the company’s history. Sonnet 5 at these prices is a growth move. It lowers the marginal cost of every developer who wants to try a coding agent for real work, and it does it right as the market is asking whether agentic coding is a real product category or an expensive habit.
The next quarter will show whether the price cut widens the market faster than it compresses Anthropic’s margins. Either way, the ceiling on what a mid-tier agent model costs to run just dropped, and everyone else’s pricing page now has to answer for that.