Claude Sonnet 5.5: The Free Claude Now Does Paid-Tier Work

The short version: Anthropic released Claude Sonnet 5.5 on September 28, and it is now the Sonnet model behind Claude’s free plan. It costs you nothing. What changed is the work it can do: on GDPval-AA, which Anthropic describes as a test of real-world work across a variety of occupations, it scored 1844 against Sonnet 5’s 1449, and on OSWorld 2.1, which grades an AI operating ordinary desktop software, it went from 57.0 percent to 80.1 percent. The free Claude app runs it at Medium effort, and Anthropic says that even at Medium it clears Sonnet 5’s best score.

That last detail is easy to miss, and it is the one that matters if you use Claude in a browser rather than through a developer’s code. Launch tables often show a model at its best setting; Anthropic’s own coding row is labeled Max effort. The Claude apps do not run at Max. Anthropic’s own launch post answers the obvious question directly: at Medium, the default in the Claude apps, Sonnet 5.5 “far exceeds Sonnet 5’s best score” for less than a tenth of the cost per task. Put plainly, the setting you get without touching anything is already better than the old model at full stretch.

This is good news with no catch attached, which is rarer than it should be. Here is what the free Claude Sonnet 5.5 upgrade actually changes for a small business, and where it still falls short.

Is Claude Sonnet 5.5 free?

Yes. Claude’s pricing page, read on September 29, lists the Free plan at $0 with Sonnet and Haiku models, and Simon Willison, the independent developer who writes up major releases on launch day, noted on September 28 that Sonnet 5.5 is now the model used for the free tier on claude.ai. The Free plan includes web search, file creation, code execution, memory, artifacts, up to five projects, skills and connectors, on the web, desktop and phone.

What Free does not include, per the same page: Claude Code, the Research feature, and the larger Opus and Fable models. Usage on Free is limited but the page does not publish a number; Pro promises at least five times the Free allowance per five-hour session. Pro costs $20 a month, or $17 a month billed annually ($200 up front), both verified September 29.

For developers and the software vendors you rent from, the API price is unchanged from Sonnet 5 at $2 per million input tokens and $10 per million output tokens, with a 1 million token context window. Anthropic says it typically finishes a job in fewer tokens, which is where its “up to 30% less per task” claim comes from.

What can the free Claude do now that it could not do last week?

Anthropic’s launch post publishes a long benchmark table, most of it about coding. Four rows describe work an owner actually does: finishing office tasks, operating software, reading charts and answering hard questions with research tools.

The chart row is the one to look at twice. Chartography is not a test of reading a bar chart; it is 100 tasks built from the specialized charts professionals make decisions from, written by people who work with those charts for a living. When its authors published it in August, the best configuration they tested reached 45.0 percent. Sonnet 5.5 posts 61.6 percent without tools, up from 15.6 percent for the model it replaces. For an owner, that is the difference between an assistant that guesses at the usage graph on a utility bill or the trend line in a supplier’s price sheet and one that mostly reads it correctly.

Which version of Sonnet 5.5 are you actually using?

Claude models now have an effort dial with five settings: low, medium, high, xhigh and max. Higher effort means the model reasons longer before answering. Anthropic set the default to Medium in the Claude apps and Claude Code, and to High for developers calling the API.

More effort is not automatically better, and Willison’s launch-day test shows why. He runs the same drawing prompt on every model. At max effort, Sonnet 5.5 thought for 128,000 tokens, spent $1.28 of API credit, ran out of room and produced nothing. At xhigh, it finished in 41 seconds for 5.74 cents. If a tool you pay for lets you pick the effort level, the top setting is for the rare hard problem, not a default.

If your help desk, CRM or scheduling software runs on Claude, the upgrade may reach you without your doing anything. Zendesk’s Director of AI, Abhinay Kathuria, said in Anthropic’s launch post that in its testing tickets were processed 20 percent faster.

What does this mean for a five-person business?

When Sonnet 5 became the free default in July, we wrote that the free tier had become worth a real look. Sonnet 5.5 moves the line again. Epic Games’ chief operating officer, Daniel Vogel, said the model cleared “the same quality bar you’d expect from a higher-tier model” in Epic’s reviews, and app builder Base44 reported that across 118 real builds its apps scored level with Opus 5, which was a paid-plan model.

So the question for a small team changes from “who needs the paid plan to get a good model?” to “who needs more usage, Research, or Claude Code?” Someone drafting customer replies, reading a quote before it goes out, or making sense of a supplier’s chart can do that well on Free. The person who hits the usage limit every afternoon, or who needs Research for a bid, is the one who needs a $20 seat. One Pro seat and four Free accounts is $20 a month; five Pro seats is $100.

One caution that has nothing to do with the model: Free and Pro are personal plans. If staff will paste customer records into Claude, look at the Team plan, which starts at $25 a seat monthly or $20 billed annually, adds central billing and admin controls, and runs under Anthropic’s commercial terms. It also sits in our guide to AI tools by the job you need done.

Where does Sonnet 5.5 still fall short?

An 80.1 percent score on desktop tasks means roughly one task in five still does not get done. A 61.6 percent score on professional charts means close to four in ten still go wrong. Those are large improvements and they are not a reason to stop checking. Anything that goes to a customer, a bank or a tax form still gets a human read.

Willison also compared the free offerings directly and concluded that Anthropic currently offers a more capable free tier than ChatGPT does. That lead may not last; Anthropic says a Haiku 5.5 model joins the family in the coming weeks, and every lab is shipping on a cycle measured in weeks. For more on how quickly these tiers move, see our look at what Grok 4.7 costs on a real month of customer replies.

The practical takeaway is simple. If your team tried the free Claude earlier this year and decided it was not good enough, it is worth one more afternoon.

Frequently Asked Questions

Is Claude Sonnet 5.5 available on the free plan?

Yes. Sonnet 5.5 is now the Sonnet model on Claude’s free plan on web, desktop and mobile. The Free plan has usage limits Anthropic does not publish as a number, and it excludes Claude Code, Research, and the larger Opus and Fable models.

What effort level does the free Claude app use?

The Claude apps run Sonnet 5.5 at Medium effort by default, while the API defaults to High. Anthropic says Sonnet 5.5 at Medium still exceeds Sonnet 5’s best benchmark score for less than a tenth of the cost per task.

Did the API price for Claude Sonnet 5.5 change?

No. It is $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5, with a 1 million token context window. Anthropic says it usually needs fewer tokens to finish a task, so the cost per job can drop by up to 30 percent depending on the work.

Should a small business upgrade everyone to Claude Pro?

Usually not. Sonnet 5.5 on Free handles everyday drafting, reading and chart work well. Pro at $20 a month makes sense for the people who hit usage limits daily or need Research, Claude Code or the Opus models, and the Team plan is the better fit if staff will handle customer data.

Have you tried the free Claude since the upgrade? Tell us the first real task you gave it and whether it got it right.