Claude Sonnet 5 just shipped.
By John Aspinall ✱ · July 1, 2026 · Curated by George's Blog
Claude Sonnet 5 just shipped.
Most operators will scroll past it.
⤷ That's the actual problem.
You're not paying per token.
You're on Claude Pro, Max, or Claude Code.
Same subscription, same price, every month.
Who cares, right?
When a new model drops, you're not deciding whether to spend more.
You're deciding whether to keep using the worse version of something you already paid for.
That's a different mistake than a pricing mistake.
→ It's a laziness mistake.
Here's what changed:
🟠 Near-Opus output quality
⤷ Inside the tier most people already have access to.
🟠 1M token context as the default
⤷ Not something you had to configure.
🟠 Adaptive thinking
⤷ Stops burning your usage on problems that didn't need deep reasoning.
That last one matters more than people realize.
If you're on a capped plan, the model wasting effort on easy tasks is the same as it eating your quota for nothing. A model that thinks adaptively gives you more real output before you hit a wall.
You already paid for this upgrade. It shipped today.
Most people won't notice for months.
I see this constantly.
Operators running the same model in Claude Code, the same default in their agent setup, the same workflow, for half a year, because nobody went and changed the dropdown.
The subscription isn't the moat.
Checking what your subscription actually gets you, the week it changes, is.