Readers who build with Claude get a faster model at the same listed API price. Claude Sonnet 5.5 responds more than 30% faster than its predecessor and uses fewer tokens on many tasks, which lowers the cost per task without changing the published price. The release is aimed at coding and office work, and it is available with zero data retention through Anthropic, Amazon Web Services, Google Cloud, and Microsoft Azure.<\/p>\n
Two numbers frame the upgrade. On Terminal-Bench 4.0, an agentic coding test that measures how well a model completes complex, multi-step professional tasks inside a command-line interface, Sonnet 5.5 scored 70.6%. Sonnet 5 scored 10.3%. On GDPval-AA, a test that covers 44 occupations across nine industries, the new model scored 1,844, two points below Opus 5.5 and 395 points above Sonnet 5.<\/p>\n
Speed and cost moved together. At the same High effort setting, the model scored about 10 points higher than Sonnet 5 on the FrontierCode coding test at roughly one-fifteenth the cost per task. At lower effort settings it is cheaper per task than Opus 5.5, and at higher settings it can match Opus 5.5 on several benchmarks at a similar cost.<\/p>\n
Users can pick an effort level to balance cost, speed, and quality. The default is Medium inside Claude Code and Anthropic’s apps, and High on the Claude Platform. Lowering the setting trims spend; raising it can lift benchmark scores.<\/p>\n
Developers select the new model on the Claude Platform using the model name Early users described the new model as quick to read a codebase and able to group tool calls into fewer steps. Reports from production teams pointed to better judgment across tasks of different complexity, fewer output tokens, and less reaching for web search. One engineering team said the drop in unnecessary web searches and the lower token use were enough to start moving simple and moderate code reviews over right away, with more to follow.<\/p>\n For office work, users said the model felt like a more natural conversation partner. It followed slide templates closely enough to produce decks that needed little editing. On AA-Briefcase, a test of long-running knowledge work, Sonnet 5.5 scored above Sonnet 5 and above GPT-5 Sol. It also closed in on Opus 5.5 on computer-use and chart-recognition tests.<\/p>\n An automated audit covering about 1,850 scenarios found that Sonnet 5.5 matched or improved on Sonnet 5 across most alignment, resistance to misuse, and honesty checks. The tests did not find evidence of the model pursuing goals that conflict with user intent. Anthropic notes its evaluations cannot catch every failure mode.<\/p>\n Cybersecurity safeguards ship with the model. Routine software development, including finding and fixing bugs, stays available. Higher-risk cybersecurity work falls back to Sonnet 5. The same biology safeguards from Sonnet 5 carry over, which can flag some legitimate microbiology and virology requests; organizations can apply to Anthropic’s verification programs for expanded access.<\/p>\n To resist attempts to extract capabilities through large numbers of fake accounts, Anthropic has added new protective measures. Most developers will not see a change in day-to-day use. Teams that move conversations between accounts should review Anthropic’s published guidance before migrating.<\/p>\n Three things are true at once. The same API price buys a faster model. The same high-quality output on coding and office tasks costs a fraction of what it did before. And the safety surface is broader, with higher-risk cybersecurity work routed to the older model. For teams already running Sonnet, the path forward is a model-name swap and a one-line setting change.<\/p>\n Yes. Anthropic says it responds more than 30% faster than Sonnet 5 and uses fewer tokens on many tasks, lowering the cost per task while keeping the same listed API price.<\/p>\n Use Users who run Sonnet with thinking turned off must switch to the new claude-sonnet-5-5<\/code>. Users who run Sonnet with thinking turned off must switch to the new between_tools<\/code> setting before moving to Sonnet 5.5. Anthropic’s release notes call this out as a required change, not an optional one.<\/p>\nReal-world feedback on coding and office work<\/h2>\n
Safety measures and what is locked down<\/h2>\n
What the upgrade means in practice<\/h2>\n
FAQ<\/h2>\n
Is Claude Sonnet 5.5 faster than Sonnet 5?<\/h3>\n
What model name should developers use for Sonnet 5.5?<\/h3>\n
claude-sonnet-5-5<\/code> when selecting the model on the Claude Platform. It is also available with zero data retention through Amazon Web Services, Google Cloud, and Microsoft Azure.<\/p>\nDo users need to change settings before switching to Sonnet 5.5?<\/h3>\n
between_tools<\/code> setting before moving to Sonnet 5.5. Higher-risk cybersecurity work will fall back to Sonnet 5.<\/p>\n