Skip to main content
VluxTech
Blog · AI

Claude Sonnet 5.5: same price, faster. What it means for you

On 28 September 2026, Anthropic released Claude Sonnet 5.5. It's an AI model, the "brain" behind a tool like a chatbot. It costs the same per use as the version it replaces, and Anthropic says it's faster and needs fewer steps.

The short verdict: if you use Claude, or a tool built on it, there's nothing to set up. If you pay for a chatbot built on Claude, it's worth asking your provider one question.

What changed

Sonnet is the middle model in Claude's lineup. It costs less than Opus, the top one. Anthropic says Sonnet 5.5 is:

  • Faster. It writes answers over 30% faster than Sonnet 5, the version before it.
  • Cheaper to finish a job. A job can cost up to 30% less, since it needs fewer tokens (small pieces of words) and tool calls. A tool call is a step where the AI uses another program.
  • The same price per token. More on that below.

It's live in Claude, in Anthropic's developer service, and on Amazon, Google and Microsoft's cloud services. Anthropic says Claude Haiku 5.5, a model for high-volume, low-cost uses, will follow "in the coming weeks." Source

How the price compares

Developers pay for AI by the token, a small piece of a word. These prices are per million tokens, for text going in and text coming out.

  • Sonnet 5.5: $2 in, $10 out. That's the same as Sonnet 5. Source
  • Opus 5.5: $4 in, $20 out. So Sonnet 5.5 costs half as much per token. Source
  • OpenAI's GPT-6 Sol: $2 in, $10 out, the same as Sonnet 5.5. Source

The same price per token doesn't mean the same bill. If Sonnet 5.5 uses fewer tokens for a job, you pay for fewer. That's where "up to 30%" comes from. Treat it as a best case.

How the performance compares

Anthropic published test scores on its launch page. Outside groups ran some of the tests, but we haven't found an independent check yet. These are the ones that matter for everyday business use:

  • Office work. On GDPval-AA, a knowledge-work test, Sonnet 5.5 scores 1844. Opus 5.5 scores 1846, and Sonnet 5 scored 1449. That's nearly level with Opus. Source
  • Reading charts. On the Chartography test, Sonnet 5.5 scores 61.6%. Sonnet 5 scored 15.6%, and Opus 5.5 scores 64.4%.
  • Hard coding. Here Opus still leads. On FrontierCode, Sonnet 5.5 scores 46.2% and Opus 5.5 scores 54.4%.

GPT-6 Sol scores 1487 on the office-work test. Anthropic's footnote says OpenAI recently fixed a bug, so that score may be out of date.

Some customers shared results too. Zendesk, which makes customer-support software, says tickets were processed 20% faster. Box says Sonnet 5.5 was more accurate, 2.4 times faster and used 12% fewer tokens than the last model. These are the companies' own results.

Why it matters for a small business

  • You pay for a Claude subscription. Your price stays the same. Sonnet 5.5 may appear in the model menu if your plan includes it.
  • You use a chatbot or tool built on Claude. The company behind it pays per token, so its cost per job may fall. Replies may also arrive faster. Some providers pass savings on, and some don't.
  • You're planning a chatbot with a developer. Sonnet 5.5 costs half as much as Opus 5.5 per token and scores about the same on office work. That makes it a sensible model to test first.

What to do now

  1. Check your model menu. Open the model menu in Claude. If Sonnet 5.5 is there, try it.
  2. Ask your provider. You can copy this message:

Hi, Anthropic released Claude Sonnet 5.5 on 28 September. Does our chatbot use it yet, or will it soon? And will the lower cost change what we pay?

  1. Test before you switch. Send your five most common customer questions, like "Are you open on Sunday?", to the old and new versions. Compare the speed, and check that prices, hours and policies are still right.
  2. Building it yourself? Tell your developer the model name is claude-sonnet-5-5. For very simple, high-volume chat, it may be worth waiting for Haiku 5.5 before you switch everything.

The catch

  • Most numbers are Anthropic's own. It doesn't say how it measured the speed or the cost saving. VentureBeat asked, and hadn't had an answer when it published. Source
  • "Up to 30%" is a best case. The price per token didn't change.
  • Opus is still better at the hardest work, like tough coding.
  • Some security-related requests go to the older model. Anthropic says higher-risk cybersecurity tasks fall back to Sonnet 5. Routine bug fixing is not affected.
  • Not confirmed: which Claude plans include it, and which countries get it, Indonesia included.

Wondering if an AI assistant for customer messages fits your business? See how we set up business automation, or book a free consultation.

Claude Sonnet 5.5: same price, faster. What it means for you · VluxTech