Why Anthropic Just Made Its Smartest AI Nearly Free
Edition 273 — The pricing move, the catch nobody's talking about, and who actually saves money.
Anthropic just did something companies almost never do: they took their best new model and made it cheaper than the one it's replacing.
Claude Sonnet 5 became the default model for every Free and Pro Claude user on July 1. Not a downgrade dressed up as an upgrade — an actual jump. Early testers are describing agent tasks that used to stall halfway now completing end-to-end, and reliability gains that show up in production work, not just benchmark charts.
Here's the part that should get your attention: it performs close to Anthropic's flagship Opus tier on a lot of real tasks, and through the end of August it's priced below the model it's replacing.
Why give away the good stuff?
The obvious read is a genuine strategy shift, not charity. Businesses spent the first half of the year getting burned by "agentic" AI bills that spiraled — some engineering teams were quietly running up thousands of dollars a week in AI usage with nobody watching the meter. That's not a growth story, that's a budget-office horror story, and it was starting to make companies gun-shy about adopting AI tools at all.
A cheaper, more reliable model at the same subscription price solves that problem for a huge chunk of users without anyone having to build a new billing dashboard. Reliability is also the more expensive thing to buy — it's a lot easier to drop a price than to make an agent stop losing the thread halfway through a task.
What this actually means for you
If you're on Pro, Max, Team, or Enterprise, you already have it — Sonnet 5 quietly replaced Sonnet 4.6 as your default on July 1, at no extra charge. If you're on the free tier, you got the same automatic upgrade, at the $0 you were already paying. Either way: same price you're used to, a meaningfully more capable model underneath it.
The one group actually paying less: developers and businesses billed by the API token. Sonnet 5's API rate — $2 per million input tokens, $10 per million output, through August 31 — undercuts Sonnet 4.6's standing $3/$15 rate. That's a genuine, if temporary, price cut, and it's worth knowing if that's how your team pays. (One catch on that below.)
The practical upgrade is in how much you can hand off. Multi-step tasks — pulling data, reformatting it, drafting a summary, checking it against a source — are the category that benefits most from a model that stays on-plan instead of drifting after step two.
| Claude Sonnet 5 | Claude Opus 4.8 | |
|---|---|---|
| Best for | Daily work, multi-step tasks, coding | Hardest reasoning problems |
| Relative capability | Near-flagship on most tasks | Highest overall |
| Intro API pricing (through Aug 31) | Lower than the model it replaces | Premium tier pricing |
| Available to | Free, Pro, Max, Team, Enterprise, API | Max, Team, Enterprise, API |
For almost everything on your daily task list, Sonnet 5 is the right tool. Save reaching for the top-tier model for the handful of problems that genuinely need the extra reasoning depth.
The one thing worth knowing before you build on it
Sonnet 5 uses a new tokenizer that counts text slightly differently than before — it can produce more tokens for the same piece of writing. That's invisible if you're chatting in the app. It matters if you're running anything at scale through the API, because "per-token" pricing now measures slightly different tokens.
The bigger picture: the AI-model price war that's been building all year just got a very public data point. When the company known for charging premium prices for frontier intelligence makes its newest model cheaper than the last one, that's a signal about where this market is heading — and it's good news if you're the one paying the bill.