Anthropic launched Claude Sonnet 5 on June 30, 2026: performance close to Opus 4.8 at $2/$10 per million tokens (launch pricing through August 31), and it's the default model for Free and Pro plans.
Ever kicked off an AI project, watched it go great for half an hour, then glanced at the spend counter and felt your mouth go dry? That little jolt of "uh oh, this is getting out of hand" is about as universal as it gets when you build with frontier models. Well, on June 30 Anthropic dropped Claude Sonnet 5 and made it — no permission asked, no extra charge — the default model across every Free and Pro plan. Overnight, millions of people woke up to a model that can make plans, open the browser, use the terminal and run multi-step tasks on its own — without touching their wallet. When the "good model" becomes the baseline model, the whole conversation changes.
What makes it special
Anthropic calls it the most agentic Sonnet they've ever built, and it isn't empty marketing. Where earlier Sonnets would stall halfway through a complex task and leave you holding the mess, this one finishes the job and reviews its own work without being asked. On performance it gets dangerously close to Opus 4.8 — the family's big, pricey sibling — but for a fraction of the cost. Add fewer hallucinations and better refusals when faced with malicious requests, and you get a model that isn't just smarter, it's more trustworthy. It's the kind of leap you won't notice in a two-minute demo, but that you'll deeply appreciate three hours into a real problem.
The number that actually matters
Here's the part that makes those of us who build with AI raise an eyebrow: it ships with launch pricing of $2 per million input tokens and $10 for output through August 31, 2026 (after that it rises to 3/15). That price is essentially cost-neutral compared to the previous Sonnet, which means you get more capability without the bill blowing up. And note the detail: this isn't a pricey model you'll "someday" get around to trying — they made it the baseline for half the internet from minute zero. When a default model becomes both more capable and cheaper at once, it isn't just another update — it's one of those things that moves the needle for the entire market.
What it means if you build on NeuralOS
Cheaper, more agentic models are literally a tailwind for a platform like NeuralOS, where behind every app you build there are tokens quietly running. When the default model for half the planet knows how to plan and use tools on its own, building AI products stops being a luxury reserved for whoever has budget to spare. But we'll tell it to you straight: we don't marry any single model. We stay model-agnostic on purpose, so if tomorrow another provider ships something better or cheaper, you swap engines with one click, not by rewriting your product. The underlying lesson is the same as always: the edge isn't in the model of the moment, it's in not depending on any of them.