Anthropic released Claude Sonnet 5 on June 30, and the pitch is unusual: it's the mid-tier model, not the flagship, but it does work that used to require the flagship. Sonnet 5 performs close to Opus 4.8 — Anthropic's heavyweight — on the things businesses actually use AI for, at a materially lower price. If you run any kind of repetitive AI task, this is the release worth paying attention to, because it changes the math on what you can afford to automate.

What actually changed

Sonnet 5 is built to be the most "agentic" Sonnet yet — meaning it can plan a multi-step task, use tools like a browser or a terminal, and keep going without stalling halfway, which older mid-tier models tended to do. It ships with a one-million-token context window by default and lands near Opus 4.8 on reasoning, coding, and knowledge work. Introductory pricing is $2 per million input tokens and $10 per million output tokens through August 31, then $3 and $15 after. One quirk: it uses a new tokenizer, so the same text counts as roughly 30% more tokens than before — factor that into any cost comparison.

The tier question, in plain terms

Anthropic now sells a ladder: fast-and-cheap (Haiku), balanced (Sonnet), and top-end (Opus 4.8 and the even larger Fable 5). The old rule of thumb was "use the big model when quality matters, the cheap one when it doesn't." Sonnet 5 blurs that. For a lot of real business work — drafting, summarizing, running a defined multi-step workflow — the balanced model is now good enough that reaching for the flagship is often paying extra for headroom you won't use.

How this stacks up against the field

OpenAI and Google both sell the same three-tier idea, and their mid-range models are competitive — the differences at this level are smaller than the marketing suggests. The honest advice: don't switch platforms over one release. If your data and tools already live in one ecosystem, the cost of moving usually outweighs a modest quality edge. The bigger news is directional — mid-tier models everywhere are getting flagship-adjacent, and prices keep falling.

The takeaway

If you're running AI workflows on a top-tier model out of habit, test the same job on Sonnet 5 and compare the output side by side. There's a real chance you get near-identical quality at a fraction of the cost. And remember the tokenizer change when you do the math — measure real spend on your actual work, not the sticker price.