The news
Anthropic released Claude Sonnet 5.5 on September 28, a mid-tier model it says runs more than 30% faster than Sonnet 5 and costs up to 30% less per task, while per-token prices stay at $2 input and $10 output per million tokens.
The savings come from efficiency rather than a rate cut. Anthropic said Sonnet 5.5 typically needs far fewer tokens to finish the same work, and that it is its fastest Sonnet model yet. Cache reads remain $0.20 per million tokens and cache writes $2.50. The model is available through Anthropic's Claude Platform as well as Amazon Web Services, Google Cloud and Microsoft Azure, with a zero-data-retention option.
On Anthropic's published benchmarks, Sonnet 5.5 comes close to Opus 5.5, which launched on September 22 at $4 and $20 per million tokens. It scored 1844 on the GDPval-AA v2.1 office-work evaluation against 1846 for Opus 5.5, and 80.1% on the OSWorld 2.1 computer-use test against 81.8%. On Terminal-Bench 4.0, a test of agentic coding in a command-line environment, it scored 70.6%, above Opus 5.5's 66.4% and far ahead of Sonnet 5's 10.3%.
Customer results published by Anthropic point the same way. Zendesk said tickets were processed 20% faster in its tests, Box said the model was 2.4 times faster while using 12% fewer tokens, and Slack said its Slackbot evaluations used about 14% fewer output tokens. Anthropic said Claude Haiku 5.5, for high-volume and cost-sensitive work, will follow in the coming weeks.
Because Sonnet 5.5's cyber skills are much stronger than its predecessor's, Anthropic is deploying it with safeguards similar to Opus 5.5: higher-risk cybersecurity tasks will fall back to Sonnet 5. It is also the first Sonnet model with classifiers designed to block reasoning extraction, which Anthropic ties to distillation attacks, where attackers use large numbers of fake accounts to copy a model's capabilities. Anthropic also said its containment evaluations, which measure whether a model pushes against the boundaries of its sandboxed test environment, found Sonnet 5.5 less inclined to do so than any other Anthropic model.
The numbers
- Input price per million tokens
- $2 (unchanged)
- Output price per million tokens
- $10 (unchanged)
- Speed vs. Sonnet 5
- 30%+ faster
- Cost per task vs. Sonnet 5
- Up to 30% lower
- Terminal-Bench 4.0
- 70.6% (Opus 5.5: 66.4%)
- GDPval-AA v2.1
- 1844 (Opus 5.5: 1846)
Why CEOs should care
For technology buyers and CFOs, the practical question is which workloads still need a premium model. On Anthropic's own numbers, Sonnet 5.5 nearly matches Opus 5.5 on office tasks and computer use at half the per-token price, and beats it on one agentic coding test. Teams running document drafting, support triage, spreadsheet work or routine coding on top-tier models should rerun their own evaluations and move eligible traffic down a tier. Ask for per-task cost, not per-token price, since the savings depend on fewer tokens and tool calls.
Budget owners should also check how usage is metered. Anthropic noted that effort settings trade speed and cost against quality, with different defaults in its apps and on its developer platform. Finance teams should confirm which setting their integrations use, because the same model can produce very different bills.
CISOs and security teams should note two changes. Higher-risk cybersecurity requests will fall back to the older Sonnet 5, which could affect internal red-team or security tooling built on Claude; Anthropic says defenders will be able to apply to an expanded Cyber Verification Program. And the new anti-distillation classifiers and preserved thinking, which ties reasoning to the originating account, may change how multi-account or proxy setups behave. Test before switching production traffic.
The bigger picture
Model releases have become a cost-performance race in the middle of the market. Sonnet 5.5 arrived six days after OpenAI launched GPT-6 Sol and Luna on September 22, which TechCrunch reported OpenAI is offering at half the cost of the previous Sol and Luna versions, and VentureBeat reported that GPT-6 Sol carries the same $2 and $10 pricing as Sonnet 5.5. For buyers, that means more capability at the same list price: on Anthropic's tests, Sonnet 5.5 scores within two points of Opus 5.5 on office work and computer use at half the per-token price. That rewards organizations that route work across models rather than standardizing on one.
What’s next
Watch for Claude Haiku 5.5, which Anthropic says is weeks away, and for independent benchmark results that test the company's claims. Enterprise customers should also track when cloud marketplaces update their Sonnet 5.5 pricing and quotas, and whether Anthropic changes default effort levels as usage grows. Buyers with multi-year Claude commitments should also ask how quickly older Sonnet versions will be retired.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
