Claude Sonnet 5.5 Got Smarter and Cheaper. That’s What I Want From Every Model Release.

Last updated September 29, 2026

Anthropic released Claude Sonnet 5.5 yesterday, six days after Opus 5.5, and I’ve been using it since it came out. It’s only been a short time, so this is an early read. It’s definitely smarter than Sonnet 5, and it seems faster.

Opus 5.5 was a big jump for me. I feel like the model got smarter, it feels faster, and it doesn’t burn through my usage as fast. That’s what I’m looking for in any model release. Companies should be able to keep honing these models so each one gets smarter and cheaper to run, because we should be seeing improvements.

What’s new in Claude Sonnet 5.5

Claude Sonnet 5.5 is a faster upgrade to Sonnet 5 that gets the same work done for less, built more for everyday tasks than for the hardest problems. Anthropic’s announcement says it runs more than 30% faster and costs up to 30% less per task.

The price itself didn’t change. Developers still pay $2 per million input tokens and $10 per million output (tokens are the chunks of text a model reads and writes), which is half what Opus 5.5 costs. The savings come from Sonnet 5.5 needing fewer tokens to finish the same job, and if you pay a flat monthly fee, that should mean your usage stretches further.

Anthropic says it’s strongest at “well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets.” On a test of real work across a range of jobs, it landed just two points behind Opus 5.5. Anthropic is also upfront that Opus 5.5 “remains clearly stronger” at complex, open-ended work, so Sonnet is for the quick doc or the spreadsheet, and Opus is still for the long, messy project.

Who gets Claude Sonnet 5.5, and is it free?

Yes, you can use it free. Developer Simon Willison noticed it’s “now the model used for the free tier on claude.ai,” so anyone using Claude without paying is on Sonnet 5.5.

Willison also points out that ChatGPT’s free tier runs on Luna 5.6, and says that gives Anthropic “a much more capable free offering” right now.

Anthropic’s announcement doesn’t break access down by plan, so if you pay for Pro or Max, open the model picker in the Claude app and look for Sonnet 5.5. Developers can find it in Claude Code too, which 9to5Mac confirmed on launch day.

Does Claude Sonnet 5.5 use less of your usage?

At the default setting it should, because Anthropic says it finishes tasks with fewer tokens than Sonnet 5. Turn the effort all the way up and it uses more tokens than any model one independent tester has measured.

Effort is how hard Claude works on an answer before it replies, and in the Claude apps and Claude Code it defaults to Medium. At Medium, Anthropic says Sonnet 5.5 beats Sonnet 5’s best coding score “for less than a tenth of the cost per task.”

Artificial Analysis ran it at max effort and measured about 193,000 output tokens per task, which it calls “the highest token use we have measured.” That’s around 60% more than Opus 5.5 at the same setting. So max effort on a subscription is probably the fastest way to hit your limit, and the default is the better choice for everyday questions.

If you’re on Pro or Max, last week’s Opus 5.5 launch also raised your five-hour limits, which takes some of the sting out of the 17% Claude Code cut earlier this month.

Why I think Anthropic is heading the right way

Both Claude 5.5 models got smarter and, at their default settings, cheaper to run, and lately OpenAI feels like it’s going the other way.

For a while there I thought OpenAI was doing a really good job of it. Recently, though, the usage does seem to have decreased with their models. (In August, OpenAI put a five-hour cap back on Codex for ChatGPT Plus users, on top of the weekly one.) So I feel like OpenAI’s going in the wrong direction and Anthropic’s going in the right one. That said, OpenAI’s DevDay keynote is today at 10 a.m. Pacific, so maybe they’ll release some things.

Either way, I don’t see anybody with brand loyalty here, which is the same thing I said in my AI race scoreboard. People want the best model and the best experience, and from what I’m seeing they don’t really care what name is on the company, within reason. People still want a company they can trust.

Claude Sonnet 5.5 FAQ

If you’re on ChatGPT and curious, the free tier is the no-risk way to try Sonnet 5.5 without switching anything.

Related reading: Claude Opus 5.5: what changes if you pay | Claude Code usage limits dropped 17% | Who’s winning the AI race | GPT-6 Sol and Luna explained | New to AI? Start here