On June 30, Anthropic shipped Claude Sonnet 5 and quietly made it the default for every Free and Pro user on the planet. The one-line version: it does most of what the flagship Opus 4.8 does, for roughly a third of the money. If you run agents, that sentence is the whole ballgame.
The numbers that matter
Anthropic put Sonnet 5 at 63.2% on its agentic coding benchmark, against Opus 4.8 at 69.2% and the old Sonnet 4.6 at 58.1%. On Terminal-Bench it jumped to 80.4% — a 20-point leap over 4.6. On a graduate-level reasoning test with tools, the gap to Opus basically vanishes. Translation: when the model can use a browser and a terminal, you are paying Opus prices for a rounding error of extra accuracy.
Pricing is the actual headline. Two dollars per million input tokens, ten per million output, locked in through August 31. After that it settles at $3/$15 — the same nominal rate as Sonnet 4.6. Enterprises that watched agent bills torch a year’s budget in a fortnight now have an off-ramp, and Anthropic knows it. TechCrunch framed the launch, bluntly, as a cheaper way to run agents. That framing was the point.
The AI tool stack actually worth paying for
One email a week. The models, tools and moves that matter, stripped of hype and filtered so you don’t have to drink from the firehose. Free, and you can bail anytime.
The catch they buried
Sonnet 5 is not a drop-in swap. Three things will quietly break your integrations if you migrate on autopilot. Extended thinking is now always on and defaults to high effort, which changes your latency and token profile whether you asked for it or not. Temperature and the other sampling knobs are gone — set them and your calls error out. And the new tokenizer chews through 1.0 to 1.35 times more tokens on the same text, so any budget cap calibrated to 4.6 is now lying to you. That “cheaper” model can get more expensive than the sticker if you do not recount.
Why now
Anthropic files for an October IPO. A mass-market model that delivers frontier-adjacent agent reliability at a price enterprises will actually renew is exactly the revenue story an S-1 wants. Sonnet 5 is a great model. It is also a pitch deck with a temperature parameter removed. Both things are true.