- Start
- Sep 28, 202690% CONFIDENCEfrom the source
Claude Sonnet 5.5 Released
- Anthropic released Claude Sonnet 5.5, "the second model in the Claude 5.5 family", on 28 September 2026, six days after Claude Opus 5.5; it is "a clear upgrade over Claude Sonnet 5, runs 30%+ faster, and costs up to 30% less for most work"[1][2][3]
- Anthropic pitches it as the faster, cheaper complement to Opus 5.5, strongest at "well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets"; CNBC calls it Anthropic's second launch since Dario Amodei urged labs to slow frontier progress[1][3]
- Its cyber capabilities are comparable to Opus 5's, so it is the first Sonnet to launch with cyber safeguards and fallbacks - higher-risk cybersecurity tasks "will visibly fall back to Sonnet 5" - and the first with classifiers that block reasoning extraction by distillation attackers[1][2]
- Available the same day on the Claude Platform as
claude-sonnet-5-5and on Amazon Web Services, Google Cloud and Microsoft Azure with zero data retention, and in GitHub Copilot[1][4][6] - Claude Haiku 5.5, "built for high-volume and cost-sensitive applications, will join the Claude 5.5 family in the coming weeks"[1][2]
Notable features
- Prices stay at Sonnet 5's $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20 and cache writes at $2.50, but it needs far fewer tokens per task; output is generated more than 30% faster, "our fastest Sonnet model to date"[1][4]
- Specifications: 1M-token context window, 128K max output (300K on the Batch API in beta), adaptive thinking with High default effort on the platform and Medium in the Claude apps, text and image input, and a June 2026 reliable knowledge cutoff[1][4]
- Benchmarks Anthropic reports: Terminal-Bench 4.0 70.6% against Sonnet 5's 10.3% and Opus 5.5's 66.4%; GDPval-AA v2.1 1844 against 1449 and 1846; OSWorld 2.1 80.1% against 57.0% and 81.8%; CursorBench 4.0 55.5%[1]
- It is the first Sonnet model to beat Pokémon Red working only from screenshots, and Anthropic says that in its testing Opus 5.5 "remains clearly stronger at complex, open-ended work requiring sustained judgment"[1]
- Includes Anthropic's invisible text watermarking, added to meet rules such as the EU AI Act[5]
Benchmarks[1]
| Benchmark | Sonnet 5.5 | Sonnet 5 | Opus 5.5 | GPT-6 Sol |
|---|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 70.6% | 10.3% | 66.4%¹ | — |
| FrontierCode 1.1 (Main) (agentic coding) | 46.2% (Max²), 52.1% (Xhigh) | 42.4% | 54.4% | 49.3% |
| CursorBench 4.0 (agentic coding) | 55.5% | 34.1% | 57.8% | — |
| GDPval-AA v2.1³ (knowledge work) | 1844 | 1449 | 1846 | 1487⁴ |
| AA-Briefcase v1.1³ (knowledge work) | 1811 | 1359 | 1822 | 1483⁴ |
| Humanity's Last Exam (multidisciplinary reasoning) | 64.5% with tools | 54.9% with tools | 67.7% with tools | — |
| OSWorld 2.1 (computer use) | 80.1% partial | 57.0% partial | 81.8% partial | — |
| Chartography (visual chart recognition) | 61.6% no tools | 15.6% no tools | 64.4% no tools | 53.6%⁴ no tools |
¹ Opus 5.5 at Xhigh effort, its highest score. ² Sonnet 5.5 scores lower at Max effort than at Xhigh on FrontierCode, which penalises out-of-scope changes. ³ Run by Artificial Analysis on a pre-release deployment of Sonnet 5.5 with a since-fixed structured-output bug. ⁴ GPT-6 Sol's scores may predate OpenAI's fix of a bug that degraded its image understanding.
References 686% CONFIDENCEOverall confidence: 86%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
- [1]90%anthropic.com/claude-sonnet-5-5anthropic.com· Posted Sep 28, 2026· Starts Sep 28, 2026 ✓· 17% of score
Anthropic's announcement is headed "Claude[4] Sonnet 5.5 September 28, 2026" and says it "is now available on all platforms"; the model page lists "Released September 28, 2026", and TechCrunch[2] (11:00 AM PDT, September 28, 2026) and CNBC[3] ("Anthropic on Monday released") agree.
- [2]90%Anthropic releases Sonnet 5.5, which it calls a significantly cheaper, faster work partner | TechCrunchtechcrunch.com· Published Sep 28, 2026· Starts Sep 28, 2026· 17% of score
TechCrunch's same-day report by Lucas Ropek (11:00 AM PDT, September 28, 2026): Anthropic[1] "has released the newest version of Sonnet", about three months after Sonnet 5, with the same cyber safeguards as Fable and Opus, and a new Haiku "in the coming weeks" with no firm date.
- [3]85%Anthropic launches cheaper AI model, its second release since CEO's call for a slowdowncnbc.com· Published Sep 28, 2026· Starts Sep 28, 2026· 17% of score
CNBC: "Anthropic[1] on Monday released a new AI model called Sonnet 5.5", its second launch since Dario Amodei urged labs to slow frontier progress, at $2 and $10 per million tokens, "half the price of Opus 5.5", with Haiku 5.5 "coming soon".
- [4]90%Claude Sonnet 5.5 - Claude Platform Docsplatform.claude.com· Published Sep 28, 2026· 17% of score
Anthropic's[1] model page: claude-sonnet-5-5, 1M-token context, 128K max output (300K on the Batch API in beta), adaptive thinking with high default effort, text and images in, a June 2026 knowledge cutoff, cache writes $2.50 and reads $0.20 per million tokens, and five breaking changes from Sonnet 5.
- [5]80%Anthropic debuts Claude Sonnet 5.5 running 30% faster than the previous-generation AI model - SiliconANGLEsiliconangle.com· Published Sep 28, 2026· Starts Sep 28, 2026· 17% of score
SiliconANGLE's same-day report adds that Sonnet 5.5 carries Anthropic's[1] invisible text watermarking, added to comply with rules including the EU AI Act, and that Haiku 5.5 is planned "in the coming weeks".
Suggest a correction
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.
