Claude Sonnet 5.5 nears Opus performance while cutting task costs up to 30%

Claude Sonnet 5.5 nears Opus performance while cutting task costs up to 30%

The new model approaches Opus 5.5 performance on several benchmarks while using fewer tokens and costing up to 30% less per task than Sonnet 5.

Share

Add us on Google
by Estefano Gomez

Anthropic has launched Claude Sonnet 5.5, the second model in its Claude 5.5 family, positioning it as a faster and cheaper option for coding, everyday work and document creation.

The model generates output more than 30% faster than Sonnet 5 and costs up to 30% less per task in Anthropicā€™s testing, largely because it requires fewer tokens to complete the same work. Token pricing itself remains unchanged at $2 per million input tokens and $10 per million output tokens.

Anthropic is pitching Sonnet 5.5 as a lower-cost complement to Claude Opus 5.5. While Opus remains targeted at complex and open-ended work requiring sustained judgment, Sonnet is designed for more clearly defined tasks, bug fixing and the creation of documents, slides and spreadsheets.

The gap between the two models has narrowed significantly on some evaluations.

Advertisement

Sonnet 5.5 outperformed Opus 5.5 on Terminal-Bench 4.0, scoring 70.6% versus 66.4%. On GDPval-AA, which measures real-world knowledge work, the two models were nearly tied, with Sonnet 5.5 scoring 1,844 compared with 1,846 for Opus 5.5.

Anthropic said that at some effort settings Sonnet 5.5 performs comparably with Opus 5.5, although the company still considers Opus the stronger model for complex tasks requiring sustained judgment.

Coding is one of the largest areas of improvement. Sonnet 5.5 gained roughly 10 points over Sonnet 5 on FrontierCode at High effort while costing about one-fifteenth as much per task, according to Anthropic. Its best CursorBench score also came within about two points of Opus 5.5.

The model also showed significant gains in broader knowledge work. Anthropic said Sonnet 5.5 came close to Opus 5.5 in computer use and chart recognition while outperforming Sonnet 5 across its internal and external evaluations.

Sonnet 5.5 is also the first Sonnet model to launch with cybersecurity safeguards similar to Anthropicā€™s higher-end models after its cyber capabilities improved substantially over the previous generation. Higher-risk cybersecurity requests may fall back to Sonnet 5, while routine development tasks remain available.

The model is available across Anthropicā€™s platforms as well as Amazon Web Services, Google Cloud and Microsoft Azure. Developers can access it through the Claude Platform using claude-sonnet-5-5.

Anthropic said Claude Haiku 5.5, aimed at high-volume and cost-sensitive applications, will join the Claude 5.5 family in the coming weeks.

Disclosure: This article was edited by Estefano Gomez. For more information on how we create and review content, see our Editorial Policy.
AI
Claude Sonnet 5.5 nears Opus performance while cutting task costs up to 30%
Claude Sonnet 5.5 nears Opus performance while cutting task costs up to 30%

The new model approaches Opus 5.5 performance on several benchmarks while using fewer tokens and costing up to 30% less per task than Sonnet 5.

by Estefano Gomez

Share

Add us on Google

Anthropic has launched Claude Sonnet 5.5, the second model in its Claude 5.5 family, positioning it as a faster and cheaper option for coding, everyday work and document creation.

The model generates output more than 30% faster than Sonnet 5 and costs up to 30% less per task in Anthropicā€™s testing, largely because it requires fewer tokens to complete the same work. Token pricing itself remains unchanged at $2 per million input tokens and $10 per million output tokens.

Anthropic is pitching Sonnet 5.5 as a lower-cost complement to Claude Opus 5.5. While Opus remains targeted at complex and open-ended work requiring sustained judgment, Sonnet is designed for more clearly defined tasks, bug fixing and the creation of documents, slides and spreadsheets.

The gap between the two models has narrowed significantly on some evaluations.

Advertisement

Sonnet 5.5 outperformed Opus 5.5 on Terminal-Bench 4.0, scoring 70.6% versus 66.4%. On GDPval-AA, which measures real-world knowledge work, the two models were nearly tied, with Sonnet 5.5 scoring 1,844 compared with 1,846 for Opus 5.5.

Anthropic said that at some effort settings Sonnet 5.5 performs comparably with Opus 5.5, although the company still considers Opus the stronger model for complex tasks requiring sustained judgment.

Coding is one of the largest areas of improvement. Sonnet 5.5 gained roughly 10 points over Sonnet 5 on FrontierCode at High effort while costing about one-fifteenth as much per task, according to Anthropic. Its best CursorBench score also came within about two points of Opus 5.5.

The model also showed significant gains in broader knowledge work. Anthropic said Sonnet 5.5 came close to Opus 5.5 in computer use and chart recognition while outperforming Sonnet 5 across its internal and external evaluations.

Sonnet 5.5 is also the first Sonnet model to launch with cybersecurity safeguards similar to Anthropicā€™s higher-end models after its cyber capabilities improved substantially over the previous generation. Higher-risk cybersecurity requests may fall back to Sonnet 5, while routine development tasks remain available.

The model is available across Anthropicā€™s platforms as well as Amazon Web Services, Google Cloud and Microsoft Azure. Developers can access it through the Claude Platform using claude-sonnet-5-5.

Anthropic said Claude Haiku 5.5, aimed at high-volume and cost-sensitive applications, will join the Claude 5.5 family in the coming weeks.

Disclosure: This article was edited by Estefano Gomez. For more information on how we create and review content, see our Editorial Policy.