Anthropic released Claude Sonnet 5.5 today, the second model in its Claude 5.5 family after Claude Opus 5.5. It generates output more than 30 percent faster than Claude Sonnet 5 and costs up to 30 percent less per task at the same per-token price.
Anthropic describes Claude Sonnet 5.5 as the faster, lower-cost partner to Claude Opus 5.5. It is strongest at well-scoped everyday tasks, bug fixes, and polished documents, slides, and spreadsheets. Claude Opus 5.5 remains the stronger model for complex, open-ended work that needs sustained judgment. Claude Haiku 5.5, built for high-volume and cost-sensitive work, is due in the coming weeks.
Claude Sonnet 5.5 Keeps the Same Token Prices and Cuts the Cost of Each Task
Pricing stays at $2 per million input tokens and $10 per million output tokens, the same as Claude Sonnet 5. The savings come from token use, since Claude Sonnet 5.5 needs far fewer tokens to finish the same work, which Anthropic says brings the cost per task down by up to 30 percent.
Against Claude Opus 5.5, the price sheet looks like this:
- Input tokens: $2 per million, against $4 for Claude Opus 5.5
- Output tokens: $10 per million, against $20
- Cache reads: $0.20 per million, the same as Claude Opus 5.5
- Cache writes: $2.50 per million, against $5
Among early testers, Balyasny Asset Management ran 2,441 finance tasks and measured about 121,000 tokens per answer for Claude Sonnet 5.5, where Claude Sonnet 5 used 497,000. Box reported responses 2.4 times faster with 12 percent fewer tokens, and Zendesk said support tickets were processed 20 percent faster.
Effort defaults to Medium in the Claude apps and Claude Code, and to High on the Claude Platform. On several benchmarks, Anthropic says Claude Sonnet 5.5 at Low or Medium effort beats Claude Sonnet 5’s best score for about a tenth of the cost per task.
Claude Sonnet 5.5 Beats Claude Opus 5.5 on Terminal-Bench 4.0 and Trails It Elsewhere
Coding shows the largest gain, with Claude Sonnet 5.5 scoring 70.6 percent on Terminal-Bench 4.0, an agentic coding test, up from 10.3 percent for Claude Sonnet 5. Anthropic lists Claude Opus 5.5 at 66.4 percent, measured at its highest effort setting. On the other benchmarks in Anthropic’s table, Claude Opus 5.5 stays slightly ahead:
- CursorBench 4.0: 55.5 percent, against 34.1 percent for Claude Sonnet 5 and 57.8 percent for Claude Opus 5.5
- FrontierCode 1.1: 52.1 percent at Xhigh effort, against 42.4 percent for Claude Sonnet 5 and 54.4 percent for Claude Opus 5.5
- GDPval-AA v2.1: 1844, against 1449 for Claude Sonnet 5 and 1846 for Claude Opus 5.5
- OSWorld 2.1 computer use: 80.1 percent, against 57.0 percent for Claude Sonnet 5 and 81.8 percent for Claude Opus 5.5
- Chartography chart recognition: 61.6 percent with no tools, against 15.6 percent for Claude Sonnet 5 and 64.4 percent for Claude Opus 5.5
On FrontierCode, Anthropic says Claude Sonnet 5.5 at High effort matches the best score from GPT-6 Sol for about a fifth of the cost per task. It is also the first Claude Sonnet model to beat Pokémon Red working only from screenshots.
In its own tests, Base44 ran 118 real app builds and found Claude Sonnet 5.5 matched Claude Opus 5 on quality in 3.6 iterations per build, where Claude Opus 5 needed 7.7. Unity said the model completed 90 percent of the tasks in its multi-step Unity Editor and coding benchmark, and Lovable saw about a third fewer tool calls.
In an internal test, Anthropic gave the model a public company’s earnings materials, call transcripts, and a slide template, then asked for a 10-slide operating review. Two experts judged the first draft ready to send as is.
Higher-Risk Cybersecurity Requests Fall Back to Claude Sonnet 5
Claude Sonnet 5.5 is the first Claude Sonnet model to launch with cyber safeguards, because its cybersecurity capabilities are comparable to those of Claude Opus 5. Routine bug finding and fixing works as before. Higher-risk cybersecurity tasks visibly fall back to Claude Sonnet 5, and cyberdefenders will soon be able to apply to Anthropic’s expanded Cyber Verification Program for tiered access.
Biology safeguards match Claude Sonnet 5, so some microbiology and virology requests may be flagged in error. Organizations can apply to the Life Sciences Verification Program for broader access.
Claude Sonnet 5.5 also launches with classifiers that block reasoning extraction, which Anthropic says makes it the first Claude Sonnet model built to resist distillation attacks. It expands preserved thinking, so the model’s thinking stays tied to the account that created it. Developers who run Claude Sonnet with thinking off need to switch to the new between_tools setting before upgrading, and Anthropic’s migration guide covers the steps.
Claude Sonnet 5.5 is available now on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, with zero data retention supported. Developers can call it with the model ID claude-sonnet-5-5.
Our Take
The jump from 10.3 to 70.6 percent on Terminal-Bench 4.0 is very imrpessive. Claude Sonnet 5.5 finishing ahead of Claude Opus 5.5 there, at half the token price, makes Claude Opus 5.5 hard to justify for routine coding. Anthropic itself says Claude Opus 5.5 is still clearly better at open-ended work, so I would use Claude Opus 5.5 to plan and Claude Sonnet 5.5 to build.
For everyday users, the speed and the slide-template result matter most, especially with Claude Docs and Slides now part of Claude chat. If you use Claude for security research, check the cyber fallback first, since those requests will drop to Claude Sonnet 5.
