Anthropic has launched Claude Sonnet 5.5, a major upgrade to its Sonnet AI model that focuses on faster responses, substantially stronger coding performance and greater efficiency.

Anthropic says Claude Sonnet 5.5 generates output more than 30% faster than Claude Sonnet 5 and can cost up to 30% less per task because it generally uses fewer tokens to complete the same work.

The new model is particularly focused on software development and agentic workflows, but Anthropic is also positioning it as a general-purpose model for creating documents, presentations and spreadsheets, operating computers and handling other professional tasks.

Claude Sonnet 5.5 delivers a huge coding boost

Coding is one of the biggest areas of improvement in Claude Sonnet 5.5.

On Terminal-Bench 4.0, which evaluates AI models on complex, multi-step tasks performed through a command-line interface, Sonnet 5.5 scored 70.6%, compared with 10.3% for Sonnet 5.

Anthropic’s testing also shows Sonnet 5.5 reaching 55.5% on CursorBench 4.0, compared with 34.1% for Sonnet 5. CursorBench evaluates coding agents on ambiguous, multi-file software development tasks based on real-world development sessions.

On FrontierCode 1.1, Sonnet 5.5 scored 52.1% at Xhigh effort, compared with 42.4% for Sonnet 5.

The improvements are not simply about producing better code. Anthropic says Sonnet 5.5 can also work more efficiently as an AI coding agent by batching tool calls and requiring fewer steps to complete tasks.

That could make a meaningful difference for developers using AI agents repeatedly, where both latency and token usage can add up quickly.

Sonnet 5.5 closes the gap with Opus

Claude Sonnet 5.5 is not intended to replace Claude Opus 5.5.

Anthropic positions Opus 5.5 for complex, open-ended tasks that require sustained judgment, while Sonnet 5.5 is designed for workloads where speed, efficiency and cost are more important.

However, the performance gap between the two models has narrowed considerably on some evaluations.

On GDPval-AA v2.1, which measures performance on professional knowledge-work tasks across 44 occupations and nine industries, Sonnet 5.5 scored 1,844 compared with 1,846 for Opus 5.5. Sonnet 5 scored 1,449.

Sonnet 5.5 also scored 1,811 on AA-Briefcase v1.1, compared with 1,822 for Opus 5.5 and 1,359 for Sonnet 5.

These benchmark results do not mean the two models are interchangeable. Anthropic continues to describe Opus 5.5 as its model for more demanding open-ended work, while Sonnet 5.5 is optimized for faster and more efficient execution.

The API price remains the same

Anthropic has not increased Sonnet’s API pricing despite the performance improvements.

Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens. Cache reads cost $0.20 per million tokens, while cache writes cost $2.50 per million tokens.

The potential cost reduction instead comes from efficiency.

Anthropic says Sonnet 5.5 can cost up to 30% less per task than Sonnet 5 because it can complete tasks using fewer tokens.

Developers can also control the model’s effort level. Lower effort prioritizes speed and token efficiency, while higher effort allows the model to spend more time reasoning and checking its work.

Anthropic says Claude applications and Claude Code default to Medium effort, while the Claude Platform defaults to High.

More than just a coding model

Anthropic is also emphasizing Sonnet 5.5’s broader capabilities.

The company says the model has improved at creating documents, presentations and spreadsheets, along with computer-use and visual-understanding tasks.

In one internal evaluation, Anthropic provided Sonnet 5.5 with quarterly earnings materials, call transcripts and a presentation template and asked it to create a 10-slide operating review. Two experts judged the resulting presentation ready to send without additional editing.

Sonnet 5.5 also scored 80.1% on OSWorld 2.1 with partial credit, compared with 57.0% for Sonnet 5. The evaluation measures an AI model’s ability to interact with a computer and complete tasks.

On Chartography’s visual chart-recognition evaluation, Sonnet 5.5 scored 61.6%, compared with 15.6% for Sonnet 5.

Anthropic also points to a more unusual demonstration: Sonnet 5.5 became the first Sonnet model to beat Pokémon Red using only screenshots as its visual input.

Together, these improvements reflect Anthropic’s broader push toward AI agents that can understand information, use software, write code and complete multi-step tasks rather than simply generate text.

Claude Sonnet 5.5 gets stronger cybersecurity safeguards

The performance improvements also come with additional safety measures.

Anthropic says Sonnet 5.5’s cybersecurity capabilities have improved enough to be comparable to Claude Opus 5, prompting the company to introduce cyber safeguards and model fallbacks similar to those used with its more capable models.

Routine software development and bug fixing remain supported, while higher-risk cybersecurity requests can be routed back to Sonnet 5.

Anthropic’s system card also documents substantially higher cybersecurity performance than Sonnet 5 in several evaluations. With cyber safeguards disabled, Sonnet 5.5 completed 46.1% of challenges in a 10-challenge subset of CyScenarioBench, compared with 0.7% for Sonnet 5.

Anthropic says Sonnet 5.5 does not cross any new thresholds under its Responsible Scaling Policy.

The company has also introduced additional safeguards against model distillation. Sonnet 5.5 launched with classifiers designed to prevent large-scale extraction of its reasoning capabilities, while preserved thinking is tied to the account that generated it.

Claude Sonnet 5.5 is available now

Claude Sonnet 5.5 is available through Anthropic’s platforms and through major cloud providers, including Amazon Web Services, Google Cloud and Microsoft Azure.

Developers can access it through the Claude API using the claude-sonnet-5-5 model identifier.

Anthropic also says Sonnet 5.5 is available with zero data retention, like Claude Opus 5.5 and Sonnet 5.

The company has said Claude Haiku 5.5 is coming in the following weeks, completing the Claude 5.5 model family alongside Sonnet 5.5 and Opus 5.5.

For Anthropic, Sonnet 5.5 represents more than another incremental model update. The company is trying to make a model that developers can run more frequently by combining stronger coding and agentic capabilities with faster output and lower per-task costs.

That combination could make Sonnet 5.5 particularly relevant for AI coding agents and other workflows where an AI model is expected to perform several actions rather than simply answer a single prompt.