Anthropic released Claude Sonnet 5 on June 30, and the pitch is simple: the cheaper, faster Claude can now plan a job and run tools like a web browser or a terminal on its own, the kind of hands-on work that until recently only ran well on the company's larger, pricier models.
What actually shipped
Sonnet is the faster, cheaper Claude that most people and most apps built on Claude run by default. Opus is the heavyweight you reach for, and pay more for, when a problem is genuinely hard. The most impressive self-directed behavior, where the model makes a plan and works through a multi-step task without you steering each click, has mostly lived in those Opus models. Sonnet 5 pulls it down into the cheaper tier.
By the company's own account, Sonnet 5 now performs close to Opus 4.8, its current top model, while costing far less to run. It is a clear step up from Sonnet 4.6 on reasoning, coding, tool use, and everyday knowledge work. On Anthropic's own knowledge-work measure it comes out a hair ahead of Opus 4.8, the one area where the cheaper model wins. It also takes in a very large amount of text at once, the equivalent of a stack of long documents, before it starts to lose the thread.
Where you'll see it
Sonnet 5 is available everywhere you already use Claude: it is the default model on the Free and Pro plans, and available to Max, Team, and Enterprise users, plus the developer API. It is also the new default model in Claude Code, Anthropic's coding tool. In practice, that means if you open Claude or use an app built on it, there is a good chance Sonnet 5 is now the model answering you, whether or not you picked it.
The honest catch
A "most agentic Sonnet yet" headline is easy to over-read. Sonnet 5 is not a new smartest Claude. On most of Anthropic's own benchmarks it still trails Opus 4.8, and the company says as much. The point of Sonnet 5 is to get most of the way to the top model for a fraction of the cost.
That distinction matters because price is what decides whether a capability actually reaches you. A skill that only works on the expensive model stays a demo for most people and most products. The same skill running on the cheap, default model shows up in the tools you already pay for.
Opus 4.8 is still the smarter Claude on most things, so this release is not about a new ceiling.
I put Sonnet 5 through my own tests, and honestly, I did not see a big difference from Opus. Opus still came out ahead more often than not.
Where it gets interesting is the price. At a fraction of what Opus costs, Sonnet 5 is good enough that for simple builds, like a landing page, it could easily be the one I reach for. For writing, I could not tell the two apart, so Sonnet 5 is the easy call there. Knowledge work is the place I want to test more, since it is the one area where Anthropic's own numbers put Sonnet 5 slightly ahead of Opus.
It fits the same pattern that keeps repeating: capable AI keeps getting cheaper while the bill holds flat. The cheap, default model now does work that used to need the premium one, and for most people that shows up as their everyday Claude getting better without costing more.