A free guide by New Money School
New Model
Anthropic just dropped Sonnet 5. Here are the real numbers, how it stacks up against their top model, what it costs, and the one thing that matters more than the model you pick.
Source
Straight from the announcement: Anthropic’s Claude Sonnet 5 release (June 30, 2026).
A new Claude model just landed, and the headlines make it sound like the only thing that matters. It is a big deal. But before we get into the specs, I want to say the quiet part out loud: the model is not the thing that makes your AI actually good. More on that at the end, because it is the part most people miss.
First, the rundown you came for. Here is exactly what Sonnet 5 is, what it costs, and whether you should care.
The BasicsWhat Sonnet 5 Actually Is
Anthropic makes three tiers of Claude models. Think of them as good, better, best: Haiku is the fast, cheap one, Sonnet is the everyday workhorse, and Opus is the top-of-the-line brain.
Sonnet 5 is the newest version of that middle workhorse. The headline is simple: it now performs close to Opus 4.8, which is their most powerful model, but at a much lower price. That is the whole story in one line. You get near-top-tier quality without paying top-tier rates.
Released
June 30, 2026
Compared to
Close to Opus 4.8, the top model
Context window
1 million tokens (roughly a very long book in one go)
Best at
Coding, agents, and long, multi-step work
The MoneyWhat It Costs
If you only use Claude inside the app, this part does not change your bill. Your Free, Pro, or Max plan is the same. Pricing matters when you or your team build things on the API, which is the pay-as-you-go version developers use.
Prices are quoted per million tokens. A token is a small chunk of text, roughly three quarters of a word. Here is what Sonnet 5 runs:
Standard price
$3 per million words in, $15 per million words out
Launch deal (through Aug 31, 2026)
$2 in, $10 out
For comparison, the top Opus model costs several times more per token. That is why "near-Opus quality at Sonnet prices" is the big deal here.
One honest catch
Sonnet 5 uses a new way of counting tokens, so the same task can use around 30% more tokens than the last version. The price per token is the same, but a job can still cost a little more overall. Not a dealbreaker, just do not expect it to be cheaper per task than Sonnet 4.6.
The ReceiptsIs It Actually Good
Short answer: yes, especially for real work like coding and running tasks on its own. Benchmarks are just tests the models take, but they give you a feel for the jump.
Coding
Real software tasks
On SWE-bench Verified, a test of fixing real code, Sonnet 5 scores 85.2%, up from 58% on the last Sonnet. That is a huge leap and lands it right near the top model.
Agents
Doing things on a computer
On OSWorld, a test of clicking around and finishing tasks like a person would, it scores 78.5%. Anthropic calls it their most agentic Sonnet yet, meaning it can plan, use a browser or terminal, and run on its own for longer.
Hard reasoning
The really tough stuff
On Humanity’s Last Exam, a brutal reasoning test, it hits 46.8% when it can use tools. It also thinks adaptively by default, so it spends more effort on hard questions and less on easy ones without you doing anything.
The Real PointThe Model Is Not Your Moat
Here is the part I actually care about. A new model drops every few weeks now. Chasing each one is exhausting, and it is not where the real advantage lives.
What makes your AI great is everything it knows about you: your work, your projects, your goals, the way you talk. And it blows my mind how many people keep all of that trapped inside one app. It is like training the perfect assistant for a year, and then one day they quit and everything they learned walks out the door with them.
Do the opposite. Keep that information for yourself, in simple files on your own device. This is called portable context, and it just means your notes live in plain files that any AI tool can read, no matter which model or app you use.
Do This Today
Start your second brain in 10 minutes
Make a folder on your computer. Add three plain text files: about-me (who you are, your role, your goals), how-i-write (your tone, a few real writing samples), and projects (what you are working on right now).
Now paste those into any AI before you ask for something. Watch how much better the answer gets when it actually knows you.
Once your context lives in files, you are free. If Claude ever goes down, you switch in seconds. If another tool gets better at something, you bring your files and go. You never re-explain yourself again, and you are ready for the next launch the second it drops, exactly like this one.
Bottom Line
Sonnet 5 is a genuine upgrade, near-top-tier quality at a workhorse price. Use it. But do not let the model chase distract you from the real advantage: owning your context. That is the thing that compounds.