Claude Sonnet 5.5 Is Here: Price, Benchmarks, and the Effort Setting That Decides Your Bill

Anthropic released Claude Sonnet 5.5 on September 28, 2026, the second model in its Claude 5.5 family after Opus 5.5. It keeps Sonnet 5's price of $2 per million input tokens and $10 per million output tokens. Anthropic says it runs more than 30% faster and costs up to 30% less per task. On several benchmarks it lands within a point or two of Opus 5.5, a model that costs twice as much per token.

The headline numbers hold up, with one condition that the launch post mentions only in passing. Claude Sonnet 5.5 is cheap when it runs at its default effort. Turn it up to max and Artificial Analysis finds it uses more output tokens than any model it has measured, which makes it cost more per task than Opus 5.5. This guide covers what changed and what the benchmarks do and do not show. It also covers who gets Sonnet 5.5, why Claude sometimes answers with Sonnet 5 instead, and when Opus 5.5 is still the better pick.

The Key Takeaways

• Same price, new model: Claude Sonnet 5.5 costs $2 input and $10 output per million tokens, half of Opus 5.5's $4 and $20. Sonnet 5 is now a Legacy model.
• Near-Opus scores: 70.6% on Terminal-Bench 4.0 against Opus 5.5's 66.4%, and 1844 vs 1846 on GDPval-AA. The Terminal-Bench gap is inside Anthropic's own error bars.
• The cost catch: at max effort, Artificial Analysis measured $7.60 per task, about 50% more than Sonnet 5 and more than Opus 5.5's $5.98.
• Who gets it: every platform on day one, model ID claude-sonnet-5-5. In Claude Code the default model stays Opus 5.5.
• New safeguards: the first Sonnet with cyber fallbacks. Flagged requests re-run on Sonnet 5, and a toggle in Settings turns switching off.

What Claude Sonnet 5.5 Is

Claude Sonnet 5.5 is Anthropic's mid-tier model, built as a faster, lower-cost complement to Claude Opus 5.5. Anthropic's launch post draws the line between the two. Opus 5.5 "is built for complex work requiring careful judgment." Sonnet 5.5 "is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets." It replaces Claude Sonnet 5, which shipped on June 30, 2026 and which we covered in our Claude Sonnet 5 launch guide.

Anthropic announced Claude Sonnet 5.5 in a thread on its official account:

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family.

It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

— Claude (@claudeai) September 28, 2026

The specs that matter

Claude Sonnet 5.5 has the same context window and output limit as Opus 5.5, and it is the fastest Sonnet Anthropic has shipped. These are the figures from Anthropic's model documentation and pricing page.

Where it sits in the Claude lineup

Claude Sonnet 5.5 sits in the middle of four current Anthropic models. Claude Fable 5.1 is the top tier at $10 and $50 per million tokens. Claude Opus 5.5, launched on September 22, is the model Anthropic's docs tell most developers to start with. Sonnet 5.5 is the speed and value tier, and Claude Haiku 4.5 remains the cheapest at $1 and $5. Anthropic says a Claude Haiku 5.5 "will join the Claude 5.5 family in the coming weeks" but has not given a date.

Sonnet 5 is still available. Anthropic's pricing page now lists it among its Legacy models, at an unchanged $2 and $10. Sonnet 5 also has a new job: it answers the requests that Sonnet 5.5's cyber safeguards flag.

Claude Sonnet 5.5 Pricing: Same Rate, Different Bill

Claude Sonnet 5.5 costs exactly what Sonnet 5 costs per token: $2 per million input tokens, $10 per million output tokens, $0.20 for cache reads and $2.50 for five-minute cache writes. Anthropic's developer blog says "swapping the model ID doesn't change your per-token bill." What changes is how many tokens a task takes, and that depends on the effort setting.

Where the 30% saving comes from

Anthropic's 30% saving comes from Sonnet 5.5 using fewer tokens per task at the same effort level. The company says it "typically needs far fewer tokens to do the same work." Anthropic's charts plot each effort level against cost per task, and the savings are largest at the lower settings. According to the Sonnet 5.5 launch post, "at Medium effort, 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" on Terminal-Bench 4.0.

Early customers report the same pattern. Slack measured "about 14% fewer output tokens" on its Slackbot evals without changing any prompts. Balyasny Asset Management ran 2,441 private finance tasks and found Sonnet 5.5 used "about 121k tokens per answer, where Sonnet 5 used 497k."

What max effort costs, measured independently

At max effort, Claude Sonnet 5.5 is more expensive per task than both Sonnet 5 and Opus 5.5. Artificial Analysis ran its Intelligence Index v4.3.2 at max effort and published the cost on its Claude Sonnet 5.5 model page. Sonnet 5.5 cost $7.60 per task, against $5.98 for Opus 5.5 and $5.09 for Sonnet 5. The reason is volume: Sonnet 5.5 generated 410 million output tokens across the index, against 260 million for Opus 5.5.

Artificial Analysis called it the heaviest token use it has recorded:

To achieve its outsized performance, Claude Sonnet 5.5 (max) uses ~193k output tokens per Intelligence Index task, the most we have measured and ~7x that of GPT-6 Astra (max).

However, its lower effort settings span a broader area of Intelligence versus Output Tokens per Task tradeoffs, with low, medium, and high effort settings sitting behind GPT-6 Sol high, xhigh, and max efforts - in this area Sol provides higher performance with fewer output tokens.

— Artificial Analysis (@ArtificialAnlys) September 28, 2026

Anthropic's own developer blog gives the same warning in gentler words. "If you're tempted to use xhigh or max effort, keep in mind that Sonnet 5.5 will think longer and cost more," it says, and it suggests Opus 5.5 for those cases. The two sources agree. Inside the Claude lineup, Sonnet 5.5 is the cheap option at Medium and High effort, and it stops being one at Max. Against OpenAI the picture is tighter: Artificial Analysis finds GPT-6 Sol gets more done per output token at the lower settings.

How to pick an effort level

Leave Claude Sonnet 5.5 at its default unless a task clearly needs more. The default is Medium in the Claude apps and Claude Code, and High on the API. Anthropic's tuning advice is to start at High for general API work, Medium for well-specified agentic coding, and Medium or Low for chat. It says to use Xhigh or Max "only where your evals show a quality gain." If you want the background on how token counts turn into a bill, our guide to what AI tokens are covers it.

Lower effort has one known side effect. Anthropic's developer blog says that at Low effort Sonnet 5.5 "sometimes skips a check that exercises the change" and reports code as done without running it. If you run Sonnet 5.5 at Low in Claude Code or through the API, Anthropic recommends a system-prompt paragraph that makes the model run a real check first. This shortened version keeps its core rules and fits in a CLAUDE.md file or a system prompt:

Before you report a code change as done, run a real check that exercises
it: the project's tests, type-checker or build, or the changed command
itself. A syntax-only check, or a check command that failed to start,
does not count. If no real check is possible here, say which one you did
not run and why instead of reporting the change as done.

Claude Sonnet 5.5 Benchmarks: Close to Opus, Far Past Sonnet 5

Claude Sonnet 5.5 beats Sonnet 5 on every benchmark Anthropic published and comes within a few points of Opus 5.5 on most of them. These are Anthropic's figures from the launch post and the system card, where the standard setting is max effort unless noted.

The Terminal-Bench win is inside the error bars

Claude Sonnet 5.5 outscoring Opus 5.5 on Terminal-Bench 4.0 is a real result, but it is not a clear lead. Anthropic's system card gives a standard error of ±2.5 points for Sonnet 5.5's 70.6% and ±2.6 points for Opus 5.5's 66.4%. The 4.2-point gap is less than twice either error. Artificial Analysis ran the test independently and got the same order with a smaller gap: 64% for Sonnet 5.5, 60% for Opus 5.5. Fair reading: the two models are roughly level on terminal work, and Sonnet 5.5 gets there at half the per-token price. Our explainer on how to read AI benchmarks covers why a few points rarely settle a comparison.

The Sonnet 5 figure is the eye-catching one. Anthropic lists Sonnet 5 at 10.3% on Terminal-Bench 4.0, so the jump to 70.6% is about sixty points. Artificial Analysis measured a 50-point gain on its own run. Either way, terminal coding is where Sonnet 5.5 moved furthest.

Where Opus 5.5 still leads

Opus 5.5 keeps a clear lead on the harder coding benchmarks and on factual knowledge. It scores 89.9% on SWE-Bench Pro against Sonnet 5.5's 81.3%, and 54.4% on FrontierCode Main against 46.2%. Artificial Analysis found Sonnet 5.5 behind on its factual-knowledge test, AA-Omniscience, at 54% against 66% for Opus 5.5, though with a lower hallucination rate. Anthropic's system card sums it up as "broadly less capable than Opus 5.5 across domains." The launch post adds that Opus 5.5 "remains clearly stronger at complex, open-ended work requiring sustained judgment."

Sonnet 5.5 does win outright in a few places. The Sonnet 5.5 system card lists it ahead of Opus 5.5 on AutomationBench, 44.7% against 42.5%. Anthropic also says it is the first Sonnet model to beat Pokémon Red working only from screenshots, a test of long-horizon work and image understanding.

The independent score

Artificial Analysis ranks Claude Sonnet 5.5 second on its Intelligence Index v4.3.2, behind only Opus 5.5. At max effort Sonnet 5.5 scores 56, Opus 5.5 scores 58, and Sonnet 5 scores 38. That is an 18-point gain over Sonnet 5.

One caveat applies to the Artificial Analysis numbers, including the GDPval-AA and AA-Briefcase scores Anthropic quotes. Artificial Analysis tested a pre-release deployment that had a bug affecting structured outputs. Anthropic says the bug is fixed and expects the effect, if any, "to be small and to understate its performance."

Who Gets Claude Sonnet 5.5

Claude Sonnet 5.5 is available everywhere Anthropic sells Claude, from launch day. That covers the Claude apps on web, desktop and mobile, Claude Code, the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry, where it runs on Global Standard deployments only. Anthropic's thread closed with "Claude Sonnet 5.5 is available everywhere today."

Free and Pro plans in the Claude apps

Claude Sonnet 5.5 is available on Claude's Free plan. Anthropic's Sonnet product page says "Anyone can chat with Claude using Sonnet 5.5 on Claude.ai," and its plan matrix lists Sonnet on every plan, Free included. Whether it is also the default model is reported rather than stated by Anthropic. Tom's Guide says it "replaces Sonnet 5 as the default model for Free and Pro users." Simon Willison, who ran a test prompt on the free tier on launch day, wrote that "it's now the model used for the free tier on claude.ai." Anthropic's launch post and release note do not use the word default, unlike its Sonnet 5 announcement in June. Our guide to what Claude's free plan includes covers the usage limits that apply either way.

In the apps, Sonnet 5.5 runs at Medium effort by default, and you can change the level from the model menu next to the send button. Thinking cannot be switched off for Sonnet 5.5 in the Claude apps, according to Anthropic's help center.

Claude Sonnet 5.5 in Fello AI

Claude Sonnet 5.5 is already available in Fello AI, the Mac, iPhone and iPad app that puts Claude, GPT, Gemini and other models in one chat window. You can pick Sonnet 5.5 from the model menu without a separate Claude subscription, then send the same prompt to another model to compare the answers side by side.

Claude Code keeps Opus 5.5 as the default

Claude Code does not switch you to Claude Sonnet 5.5 automatically. Anthropic's developer blog says "the default model stays Opus 5.5, so switch with /model sonnet for well-scoped tasks." From Claude Code v2.1.284, the sonnet alias points to Sonnet 5.5, which runs at Medium effort with the full 1M-token context. Sonnet 5.5 has no fast mode.

Anthropic's developer account pitched the switch as a way to stretch plan limits:

Sonnet 5.5 is smarter, more efficient, and 30% faster than Sonnet 5. It costs up to 30% less for most work, so your Claude Code usage goes further too.

Use it for well-scoped everyday tasks like fixing bugs and quickly iterating on features.

— ClaudeDevs (@ClaudeDevs) September 28, 2026

What breaks when developers switch from Sonnet 5

Moving an API integration from Sonnet 5 to Claude Sonnet 5.5 takes more than a model ID swap. Anthropic's migration notes list five breaking changes, and three are the most likely to bite.

Thinking can no longer be disabled. A request with thinking set to disabled returns a 400 error. Use the new between_tools setting instead, which skips up-front thinking and only thinks between tool calls.

Forced tool choice is rejected. A tool_choice of type any or tool returns a 400 error. Send auto and mark the tool strict: true.

Effort levels were recalibrated. An old setting will not produce the same amount of thinking on Sonnet 5.5, so re-run your effort tests.

On the plus side, the minimum cacheable prompt drops to 512 tokens from 1,024, so shorter system prompts now qualify for caching.

Why Claude Sometimes Answers With Sonnet 5 Instead

Claude Sonnet 5.5 hands some requests to Sonnet 5 because it is the first Sonnet with cybersecurity safeguards. Anthropic says Sonnet 5.5's cyber capabilities are "comparable to Opus 5's," so it gets the same kind of protection as Anthropic's top models. When a classifier flags a higher-risk request, Claude re-runs it on Sonnet 5 in the same chat and labels the answer with the model that wrote it.

According to Anthropic's help center, two kinds of request fall back to Sonnet 5. The first is offensive security work, such as exploit generation, binary vulnerability scanning and penetration testing. Scanning your own source code for vulnerabilities still runs on Sonnet 5.5. The second is frontier AI development, a narrow set of tasks such as kernel development for certain AI accelerators.

Two other kinds of request are blocked outright with no fallback. Biology requests that could help cause serious harm are refused, as they were on Sonnet 5. Attempts to extract Claude Sonnet 5.5's internal reasoning are also blocked, such as asking it to repeat its chain of thought word for word. Asking "why did you do that?" is not affected.

Automatic switching is on by default. To turn it off, go to Settings > Capabilities and switch off Switch models when a message is flagged. With it off, a flagged request pauses the chat instead of changing model. After a fallback, the model picker stays on Sonnet 5 for the rest of that chat until you switch back yourself. On the API, fallback is off until a developer configures it.

In practice the fallback is rare. Artificial Analysis saw Sonnet 5.5 fall back in about 0.1% of tasks across its whole index. Anthropic's own Terminal-Bench 4.0 run flagged 1.2% of requests.

Claude Sonnet 5.5 vs Opus 5.5: Which to Use

Use Claude Sonnet 5.5 for well-scoped work with a clear way to check the result, and use Opus 5.5 when the task is open-ended or expensive to get wrong. Anthropic's developer blog draws the same line: Sonnet "fits best when the task has a clear spec and a way to check the result," while "for the hardest long-horizon work, an Opus model is the better choice."

What early customers found

Early customer reports back the same split between Claude Sonnet 5.5 and Opus 5.5. Base44 ran 118 real app builds and found Sonnet 5.5's apps "scored level with Opus 5" in 3.6 iterations per build, where Opus 5 took 7.7. Kevin Ngo, a creative coder at Creator quoted in Anthropic's launch post, put the division of labour neatly: "When Claude Opus 5.5 sets the architecture and general framework for a game, I would feel confident in letting Sonnet 5.5 implement it."

How to test both yourself

If you are unsure which Claude model fits a job, the practical test is to run the same prompt through both and compare. Anthropic's plan matrix lists Opus on Pro and Max but not on Free, and our Opus 5.5 guide covers which paid plans include it. The same test works across vendors: Claude Sonnet 5.5 in Fello AI sits in the same picker as GPT and Gemini.

The Verdict on Claude Sonnet 5.5

Claude Sonnet 5.5 is the model most Claude users should now spend most of their time in. It matches Opus 5.5 on terminal coding and office work, runs faster, and costs half as much per token. The one trap is effort. At its default Medium and High settings it delivers the savings Anthropic promises. At Max it burns more tokens than any model Artificial Analysis has measured and costs more per task than Opus 5.5. So keep Sonnet 5.5 at its default for everyday work, and move to Opus 5.5 rather than turning the dial to Max when a task gets hard.

Frequently Asked Questions