← Back to blog
Claude Sonnet 5.5 Review: Benchmarks, Pricing & vs Opus 5.5

Claude Sonnet 5.5 Review: Benchmarks, Pricing & vs Opus 5.5

Emma Reed

By Emma Reed

MyClaw Editorial

Claude Sonnet 5.5

Try Claude Sonnet 5.5

Start a chat with Claude Sonnet 5.5 for coding, writing, or research.

Chat with Claude Sonnet 5.5

AI Takeaway

  • What is Sonnet 5.5? Claude Sonnet 5.5 is Anthropic's new mid-tier model, released September 28, 2026 (claude-sonnet-5-5). It has a 1M-token context window and is pitched as a faster, cheaper partner to Opus 5.5.
  • How much better is Sonnet 5.5 than Sonnet 5? A lot, per Anthropic's own numbers: 55.5% vs 34.1% on CursorBench 4.0, 30%+ faster output, and up to 30% lower cost per task.
  • Should you pick Sonnet 5.5 or Opus 5.5? Use Sonnet 5.5 for well-scoped work. Anthropic says Opus 5.5 is still clearly stronger at complex, open-ended tasks.
  • How much does Sonnet 5.5 cost? Per Anthropic, $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5 and half of Opus 5.5's $4/$20.
  • Can you run Sonnet 5.5 on MyClaw? Yes. You can pick Sonnet 5.5 for a managed OpenClaw or Hermes agent on MyClaw, and switch to Opus 5.5 for harder jobs.

Sonnet 5.5 Specs: The Numbers That Matter

Sonnet 5.5 is the second model in the Claude 5.5 family, after Opus 5.5. Claude Haiku 5.5 launched on October 7, 2026. The spec sheet:

SpecClaude Sonnet 5.5
ReleasedSeptember 28, 2026
Model IDclaude-sonnet-5-5 (anthropic.claude-sonnet-5-5 on Amazon Bedrock)
Context window1M tokens
Max output128K tokens (300K on the Batch API with a beta header)
Input and outputText and images in, text out
Knowledge cutoffJune 2026
ThinkingAdaptive, with Low, Medium, High, Xhigh and Max effort
Default effortHigh on the API, Medium in the Claude apps
PlatformsClaude.ai, Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry
RetirementNot before September 28, 2027

Anthropic says anyone can chat with Sonnet 5.5 on Claude.ai, on web, iOS and Android, free users included. It hasn't published per-plan usage limits.

Introducing Claude Sonnet 5.5 \ Anthropic

Sonnet 5.5 vs Sonnet 5: A Generational Leap on Paper

The benchmark jump is the headline of any Sonnet 5.5 review. All scores below are Anthropic-reported, taken from its launch table:

BenchmarkSonnet 5.5Sonnet 5Opus 5.5GPT-6 Sol
Terminal-Bench 4.070.6%10.3%66.4% (Xhigh)—
FrontierCode 1.152.1% (Xhigh) / 46.2% (Max)42.4%54.4%49.3%
CursorBench 4.055.5%34.1%57.8%—
GDPval-AA v2.11844144918461487
AA-Briefcase v1.11811135918221483
Humanity's Last Exam (tools)64.5%54.9%67.7%—
OSWorld 2.180.1%57.0%81.8%—
Chartography61.6%15.6%64.4%53.6%

Read it with some caution: these are vendor numbers for a day-old model. Sonnet 5's 10.3% on Terminal-Bench looks unusually low, so take that gap as a rough signal rather than a precise one. On FrontierCode, Sonnet 5.5 scored lower at Max than at Xhigh. Anthropic says the extra review work at Max led to timeouts or out-of-scope edits. GDPval-AA and AA-Briefcase ran on a pre-release build with a since-fixed bug.

The efficiency story may matter more than the scores. Anthropic says Sonnet 5.5 writes output 30%+ faster than Sonnet 5. On several benchmarks, it beats Sonnet 5's best score at Low or Medium effort for about a tenth of the cost per task. For the baseline, see our Claude Sonnet 5 review.

Introducing Claude Sonnet 5 \ Anthropic

Sonnet 5.5 or Opus 5.5? Route by How Clear the Task Is

On paper, the two models are close. Sonnet 5.5 trails Opus 5.5 by two points on GDPval-AA (1844 vs 1846) and by about two on CursorBench (55.5% vs 57.8%). It's near level on OSWorld and even leads on Terminal-Bench 4.0. Opus holds a clearer edge on FrontierCode (54.4% vs 52.1%) and Humanity's Last Exam (67.7% vs 64.5%).

Anthropic doesn't oversell it: "Opus 5.5 remains clearly stronger at complex, open-ended work requiring sustained judgment." It pitches Sonnet 5.5 for well-scoped everyday tasks, bug fixes, and polished documents, slides and spreadsheets.

So route by how clear the task is:

  • Clear spec, known shape: Sonnet 5.5, often at Low or Medium effort.
  • Fuzzy goal, many decisions, high stakes: Opus 5.5.
  • Big builds: Opus 5.5 plans the architecture and Sonnet 5.5 implements it, a split one early tester described.

For the full picture of the flagship, read our Claude Opus 5.5 review.

Same Token Price, Smaller Bill per Task

Sonnet 5.5 keeps Sonnet 5's token prices:

Price per 1M tokens (per Anthropic)Sonnet 5.5Opus 5.5
Input$2$4
Output$10$20
Cache read$0.20$0.20
Cache write (5 min)$2.50$5
Batch input / output$1 / $5$2 / $10

Per Anthropic, US-only inference adds a 1.1x multiplier, and the full 1M context is billed at the standard rate.

The real saving shows up per task. Anthropic says Sonnet 5.5 needs far fewer tokens than Sonnet 5 for the same work, so it costs up to 30% less per task at identical token prices. Effort is your main cost dial: lower is faster and cheaper, higher thinks longer. Anthropic also reports that at High effort on FrontierCode, Sonnet 5.5 matches GPT-6 Sol's best score for about a fifth of the cost per task. To see how the flagships compare on price, read Opus 5.5 vs GPT-6 Sol.

Put Claude Sonnet 5.5 to work on a real task

Start a chat with Claude Sonnet 5.5 for coding, writing, or research.

Chat with Claude Sonnet 5.5

Five Breaking Changes to Fix Before You Switch

API users shouldn't just swap the model string. Anthropic lists five breaking changes for code built on Sonnet 5:

  • Thinking off is now between_tools. Switch to it before migrating if you ran Sonnet with thinking off.
  • Forced tool use returns an error.
  • Thinking blocks are tied to the model and the conversation.
  • The older computer_20251124 tool is rejected on the Claude API and Google Cloud.
  • The advisor tool rejects Opus 4.8, Opus 4.7 and Sonnet 5 as advisors.

Also watch sampling settings: non-default temperature, top_p or top_k values return a 400 error. Text between tool calls now arrives in thinking blocks, so streaming apps may go quiet until you adjust settings. Rerun your evals before moving production traffic, and start agent loops at Medium effort.

Sonnet 5.5 Review Verdict: Who Should Switch Now

Our Sonnet 5.5 review verdict: if you're on Sonnet 5, switching is an easy call once your code handles the breaking changes. If you run Opus 5.5 on routine work, test Sonnet 5.5 as the default and keep Opus for the hard parts.

Pros:

  • Large Anthropic-reported gains over Sonnet 5 at the same $2/$10 price (per Anthropic)
  • 30%+ faster output and fewer tokens per task
  • Close to Opus 5.5 on knowledge work, computer use and CursorBench
  • 1M context, and anyone can use it on Claude.ai

Cons:

  • Opus 5.5 is still clearly stronger on complex, open-ended work, per Anthropic
  • Breaking API changes mean code updates before you switch
  • Benchmarks are vendor-reported and very fresh
  • Higher-risk cybersecurity requests visibly fall back to Sonnet 5
JobBest Pick
Bug fixes and PR-sized changesSonnet 5.5
Documents, slides and spreadsheetsSonnet 5.5
High-volume chat or supportSonnet 5.5 at Low or Medium effort
Long, ambiguous agent runsOpus 5.5
Architecture and high-stakes reviewOpus 5.5

Choosing for an agent framework? Our guide to the best model for OpenClaw goes deeper.

Run Sonnet 5.5 in an Agent That Never Clocks Out

Speed and a low cost per task make Sonnet 5.5 a natural default for an agent that works all day, sorting messages, drafting documents and handling routine requests.

On MyClaw, you can pick Sonnet 5.5 as the model for a managed OpenClaw or Hermes Agent running 24/7. When a job needs deeper judgment, like a tangled research question or an architecture call, switch that agent to Claude Opus 5.5. That's the same split Anthropic recommends, and there's no server for you to maintain. Our roundup of the best AI models for Hermes shows how model choice plays out day to day.

Chat with Claude Sonnet 5.5

Launch a Sonnet 5.5 Agent on MyClaw in 3 Steps

  1. Choose your agent. Sign up for MyClaw and pick OpenClaw or Hermes Agent.
  2. Select Sonnet 5.5 and connect. Choose Claude Sonnet 5.5 as the model, keep Opus 5.5 for harder jobs, and connect the chat app you already use.
  3. Hand it a recurring task. Describe a job you repeat daily or weekly, and let the agent run it around the clock.

For the managed setup, see our guide to managed Hermes Agent hosting.

Put Claude Sonnet 5.5 to work on a real task

Start a chat with Claude Sonnet 5.5 for coding, writing, or research.

Chat with Claude Sonnet 5.5
Claude Sonnet 5.5 Review: Benchmarks, Pricing & vs Opus 5.5 | MyClaw.ai