ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›ARTICLES›CLAUDE SONNET 5.5 VS SONNET 5: WHAT CHANGED
ARTICLES · COMPARISONBEGINNER

Claude Sonnet 5.5 vs Sonnet 5: What Changed

Anthropic’s new mid-size model, out 28 Sep 2026: who should pick it over Opus 5.5, what changed from Sonnet 5, and the five changes that can break code.

READ8 min
WORDS1,528
SECTIONS6
TYPECOMPARISON
CHECKED1 OCT 26

Claude Sonnet 5.5 is Anthropic’s new middle model, released on 28 September 2026, and it replaces Sonnet 5. Pick it for everyday, well-defined work where speed matters; pick Opus 5.5 when the task is complex and needs careful judgment. Anthropic says it is faster and costs less per job than Sonnet 5. If you call Sonnet 5 from your own code, check five changes before you switch, because some old requests now fail.1

TL;DR — THE SHORT VERSION
  • New on 28 September 2026. Sonnet 5.5 is the second model in Anthropic’s 5.5 family, after Opus 5.5.
  • Sonnet 5 is now “legacy”: still available, no longer current.
  • Sonnet for everyday jobs, Opus for hard ones. That is how Anthropic itself splits them.
  • “30%+ faster” and “up to 30% less” are Anthropic’s own claims, not measurements made here.2
  • Bigger memory, newer knowledge. A 1M-token context window and knowledge up to June 2026.
  • Developers: five breaking changes. Forced tool use and turning thinking off the old way now return errors.
  • Haiku 5.5 followed on 7 October 2026.
◈ IN PLAIN TERMS

Anthropic makes Claude in sizes. Opus is the large, careful one. Sonnet is the mid-size one: quicker and cheaper, good at most everyday work. Haiku is the small, fastest one. The number is the version: 5.5 is newer than 5. So “Sonnet 5.5” means the newest mid-size Claude.

If you only use Claude in the app or website, you do not have to do anything. The rest of this page matters most if you pick models yourself or write code that calls Claude.

Sonnet 5.5 or Opus 5.5: which to pick

Anthropic draws the line itself. Opus 5.5 is for complex work that needs careful judgment. Sonnet 5.5 is for well-defined everyday tasks: fixing bugs, and making documents, slides and spreadsheets.2

How Anthropic describes the two, read at source 1 Oct 2026
If your work is…PickWhy
Clear, well-defined, and you do a lot of itSonnet 5.5Anthropic calls it “the best combination of speed and intelligence”, and its speed is rated “Fast” against Opus 5.5’s “Moderate”.
Complex, long-running, or costly to get wrongOpus 5.5Anthropic’s default recommendation: “start with Claude Opus 5.5 for most workloads.”
Not sureOpus 5.5 firstThen try Sonnet 5.5 on the same tasks and keep it where the results hold up.

Anthropic, Models overview, read at source 1 Oct 2026: Claude Sonnet 5.5 “The best combination of speed and intelligence”, comparative latency “Fast”; Claude Opus 5.5 “Moderate”; “If you’re unsure which model to use, start with Claude Opus 5.5 for most workloads.”

TAKEAWAY

Anthropic’s own default is still Opus 5.5. Sonnet 5.5 is the one to move routine work onto once you have checked it copes.

What changed from Sonnet 5

Three things you can check on Anthropic’s own pages: how much it can read at once, how recent its knowledge is, and how long it is promised to stay.

Anthropic’s published figures. Sonnet 5.5 read at source 1 Oct 2026; Sonnet 5 read at source 16 Sep 2026
Sonnet 5.5Sonnet 5
StatusCurrentLegacy (still available)
Context window1M tokens1M tokens
Max output128K tokens128K tokens
Reliable knowledge cutoffJun 2026Jan 2026
Retires no sooner than28 Sep 202730 Jun 2027

Anthropic, Models overview, read at source 1 Oct 2026: Claude Sonnet 5.5 context window “1M tokens”, max output “128K tokens”, reliable knowledge cutoff “Jun 2026”, retirement “Not sooner than September 28, 2027”; “Legacy models (still available)” include Claude Sonnet 5. Sonnet 5’s figures as read at source 16 Sep 2026 on the same page: reliable knowledge cutoff “Jan 2026”. Anthropic, Model deprecations, read at source 1 Oct 2026: claude-sonnet-5 “Active”, deprecated “N/A”, “Not sooner than June 30, 2027”. “Legacy” here means Anthropic no longer lists it as current; it has not been deprecated.

A context window is how much text the model can take in at once. One million tokens is, by Anthropic’s own estimate, roughly 555,000 words on the current tokenizer. The knowledge cutoff is how recent its training is: Sonnet 5.5 knows about events up to about June 2026, five months later than Sonnet 5.3Reasoning, 1 Oct 2026 — January 2026 to June 2026 is five months.

Anthropic’s developer notes add that Sonnet 5.5 uses the same tokenizer (the way text is split into billable pieces) as Sonnet 5, so the same text counts the same number of tokens. Thinking is on by default and adjusts itself to the task.4

The speed and cost claim

Anthropic says Sonnet 5.5 “runs 30%+ faster, and costs up to 30% less for most work” than Sonnet 5. That is the company’s claim. This site has not timed or billed either model.2

The saving is not a lower price tag. Anthropic’s developer notes say the per-token prices are the same as Sonnet 5’s. So any saving has to come from the model using fewer tokens to finish a job. How much you save depends on your work, and “up to” means some work will save less.4Reasoning, 1 Oct 2026 — if the price per token is unchanged, a lower cost per job can only come from fewer tokens per job.

For the actual prices, see Anthropic’s pricing page. This site no longer adds new prices.

TAKEAWAY

Run a handful of your real tasks on both models and compare the bill and the result. That is the surest way to learn what Anthropic’s cost claim means for your own work.

For developers: five things that can break

If your code calls Sonnet 5 and you change only the model name to claude-sonnet-5-5, some requests will fail. Anthropic lists five breaking changes.5

  • Turning thinking off works differently. “Thinking” is the model reasoning before it answers. The old off switch (disabled) now returns an error; send between_tools instead, at high effort or below.
  • Forced tool use is gone. A “tool” is an outside function the model can call, such as a search. You can no longer force the model to call one; leave the choice on auto and say in your prompt when the tool should be used.
  • Thinking notes stay with this model and this conversation. If you edit earlier messages and resend the model’s old thinking, the request can fail. Keep conversations add-only.
  • An older computer-use tool is refused on the Claude API and Google Cloud. (Computer use lets the model operate a screen.) Move to the newer toolset.
  • Some “advisor” pairings are refused. Opus 4.8, Opus 4.7 and Sonnet 5 can no longer act as advisor to a Sonnet 5.5 model.

Anthropic, What’s new in Claude Sonnet 5.5, read at source 1 Oct 2026: “To turn off up-front thinking on Claude Sonnet 5.5, send thinking: {"type": "between_tools"} instead of "disabled"”; “Claude Sonnet 5.5 doesn’t support forced tool use.”; “Keep the conversation append-only so the check never fails”; “A request that declares the earlier computer_20251124 tool returns a 400 invalid_request_error”; “Claude Opus 4.8, Claude Opus 4.7, and Claude Sonnet 5 advisors work with a Claude Sonnet 5 executor, but with a Claude Sonnet 5.5 executor they return a 400 invalid_request_error.”

One more change fails nothing but can make an app look stuck. Short progress notes the model writes between tool calls now come back inside thinking blocks, which are empty by default. An app that shows those notes to users goes quiet until you change a display setting.4

Anthropic also says effort levels were recalibrated, so a setting carried over from Sonnet 5 will not produce the same amount of thinking. Its step-by-step guide is the Sonnet 5.5 migration guide.4

TAKEAWAY

Most of these changes match what Opus 5.5 already did a week earlier. If you migrated to Opus 5.5, the same fixes apply here.

Where you can use it

Anthropic says Sonnet 5.5 is “now available on all platforms”. Its developer docs list the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry, all under the name claude-sonnet-5-5 (Bedrock adds the prefix anthropic.).2

In Claude Code, Anthropic’s coding tool, Sonnet 5.5 became the default Sonnet model on 28 September. Choosing “Sonnet” there now gets you 5.5.6

Text from Sonnet 5.5 carries Anthropic’s hidden text watermark, according to its help centre table, read on 1 October and again on 13 October 2026. How that works is on Claude’s text watermark.7

Haiku 5.5, the small model in the same family, came out on 7 October 2026.8 When this page was first written on 1 October, Anthropic said only that it “will join the Claude 5.5 family in the coming weeks”.9

The whole current lineup is on which Claude model to use, and the Opus release a week earlier on Claude Opus 5.5 vs Opus 5.

What this page could not verify

  • The speed and cost figures. Both are Anthropic’s. Nothing was run here.
  • Benchmarks. Anthropic’s announcement reports test scores; none were re-run, so none are repeated here.
  • Which model each Claude app plan uses by default. Not stated on the pages read.
  • When Sonnet 5 will be retired. It had no deprecation notice as of 1 October 2026; “not sooner than 30 June 2027” is a floor, not a date.
  • When Haiku 5.5 would arrive. On 1 October “in the coming weeks” was all Anthropic said; it came out on 7 October (see above).
SOURCES
  1. 1Anthropic, Claude Sonnet 5.5, 28 Sep 2026, read at source 1 Oct 2026Sonnet 5.5 is “a faster, lower-cost complement to Claude Opus 5.5”. ↑
  2. 2Anthropic, Claude Sonnet 5.5, read at source 1 Oct 2026“runs 30%+ faster, and costs up to 30% less for most work”. ↑aOpus 5.5 handles “complex work requiring careful judgment”; Sonnet 5.5 is for “well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets”. ↑b“runs 30%+ faster, and costs up to 30% less for most work”. ↑c“now available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure.” Anthropic, What’s new in Claude Sonnet 5.5, read at source 1 Oct 2026, section “Availability”. ↑d
  3. 3Anthropic, Models overview, read at source 1 Oct 2026“1M tokens is roughly 555k words or 2.5M Unicode characters on the current tokenizer”; reliable knowledge cutoff is “The date through which the model’s knowledge is most extensive and reliable.” ↑
  4. 4Anthropic, What’s new in Claude Sonnet 5.5, read at source 1 Oct 2026“The tokenizer is the same as Claude Sonnet 5’s, so the same text produces the same token counts.” and “Adaptive thinking is on by default”. ↑a“Claude Sonnet 5.5 has the same prices as Claude Sonnet 5, including prompt caching and batch processing rates.” ↑b“An application that streams that text to its users goes quiet between tool calls until it sets a display value that returns the text, or turns off up-front thinking with between_tools.” ↑c“Effort levels are recalibrated. An effort level doesn’t produce the same amount of thinking as it did on Claude Sonnet 5.” ↑d
  5. 5Anthropic, Claude Platform release notes, 28 Sep 2026, read at source 1 Oct 2026“Code written for Claude Sonnet 5 can break on Claude Sonnet 5.5 in five ways.” ↑
  6. 6Anthropic, Claude Code changelog, read at source 1 Oct 20262.1.284 (September 28, 2026), “Added Claude Sonnet 5.5 (claude-sonnet-5-5), now the default Sonnet model on the Anthropic API”. ↑
  7. 7Claude Help Center, How Claude marks AI-generated content, read in a browser 1 Oct 2026 and at source 13 Oct 2026Claude Sonnet 5.5 ticked for text watermarks (in Claude output and in cloud partner output) and Content Credentials. ↑
  8. 8Anthropic, Introducing Claude Haiku 5.5, 7 Oct 2026, read at source 13 Oct 2026“Claude Haiku 5.5 is designed for high-volume, cost-sensitive tasks.” ↑
  9. 9Anthropic, Claude Sonnet 5.5, read at source 1 Oct 2026.↑

Anthropic, Claude Sonnet 5.5 (28 September 2026). Anthropic developer docs: Models overview, What’s new in Claude Sonnet 5.5, Claude Platform release notes and Pricing. Anthropic, Claude Code changelog. Claude Help Center, How Claude marks AI-generated content. All read at source on 1 October 2026; Sonnet 5’s figures as read on 16 September 2026.

Vendor claims are marketing until someone else measures them.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE