Claude Sonnet 5.5 gets faster without a price hike
Anthropic released Claude Sonnet 5.5, over 30% faster, with much stronger coding scores at the same API price.
Anthropic launched Claude Sonnet 5.5 for coding and office work, saying it responds more than 30% faster than Sonnet 5 and uses fewer tokens without changing the listed API price. It scored 70.6% on Terminal-Bench 4.0 versus 10.3% for Sonnet 5, and 1,844 on GDPval-AA, two points below Opus 5.5. The model is available as claude-sonnet-5-5 with zero data retention on AWS, Google Cloud, and Microsoft Azure. An audit of about 1,850 scenarios found alignment at least matching Sonnet 5, while higher-risk cybersecurity tasks fall back to Sonnet 5.
- Sonnet 5.5 is over 30% faster and uses fewer tokens at the same API price.
- Terminal-Bench 4.0 score rose from 10.3% on Sonnet 5 to 70.6%.
- GDPval-AA score was 1,844, two points below Opus 5.5.
- Available on AWS, Google Cloud, and Azure as claude-sonnet-5-5.
- Higher-risk cybersecurity requests fall back to Sonnet 5.
Full article524 words · extracted from helpnetsecurity.com · click to collapse
Anthropic has released Claude Sonnet 5.5, an AI model for coding and office work. The company says it responds more than 30% faster than its predecessor and uses fewer tokens to complete many tasks, reducing the cost per task without changing its listed API price.
The model is available with zero data retention and through Amazon Web Services, Google Cloud and Microsoft Azure. Developers can select it on the Claude Platform using the model name claude-sonnet-5-5. Users who run Sonnet with thinking turned off must switch to the new between_tools setting before moving to Sonnet 5.5.
Performance and cost
Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, an agentic coding test. Sonnet 5 scored 10.3%. On GDPval-AA, a test of work across 44 occupations and nine industries, the new model scored 1,844. That was two points below Opus 5.5 and 395 points above Sonnet 5.
Terminal-Bench 4.0 measures how well a model can complete complex, multi-step professional tasks within a command-line interface. (Source: Anthropic)
Sonnet 5.5 costs less per task than Opus 5.5 at lower effort settings. At higher settings, it can achieve comparable benchmark results at a similar cost. On the FrontierCode coding test, Anthropic says it scored 10 points higher than Sonnet 5 at the same High effort setting, at about one-fifteenth of the cost per task. Early testers said it understood codebases quickly and grouped tool calls into fewer steps.
“Claude Sonnet 5.5 shows better judgment than Sonnet 5 across different levels of complexity, while spending significantly fewer output tokens. Sonnet 5’s tendency to reach for web search too often and its high token use are both gone in this new model. We plan to move simple and moderate reviews over now, and more in the coming weeks,” David Loker, VP of AI at CodeRabbit, said.
Sonnet 5.5 scored close to Opus 5.5 on computer-use and chart-recognition tests. It scored above Sonnet 5 and GPT-6 Sol on AA-Briefcase, a test of long-running knowledge work. Early testers described it as a more natural conversation partner and said it could follow slide templates to create decks requiring little editing.
Users can adjust the effort setting to manage cost, speed and quality. The default is Medium in Claude Code and Anthropic’s apps, and High on the Claude Platform.
Safety measures
In an automated audit covering about 1,850 scenarios, Sonnet 5.5 matched or improved on Sonnet 5 in most measures of alignment, resistance to misuse and honesty. Anthropic found no evidence that it pursued goals conflicting with users’ intentions. The company says its tests cannot detect every failure.
The model includes cybersecurity safeguards. Users can continue routine software development tasks, including finding and fixing bugs. Higher-risk cybersecurity tasks will fall back to Sonnet 5. Sonnet 5.5 uses the same biology safeguards as Sonnet 5. Anthropic says these may flag some legitimate microbiology and virology requests. Organizations can apply to its verification programs for expanded access.
Anthropic has introduced measures to prevent attempts to extract the model’s capabilities using large numbers of fake accounts. The company says most developers will not notice a change. Those who move conversations between accounts should consult its guidance.