Key Takeaways
- Anthropic has released Claude Sonnet 5.5, a mid-tier AI model with improved coding and knowledge work capabilities.
- The new model generates responses 30% faster and can cost up to 30% less per task.
- Claude Sonnet 5.5 scores higher on coding benchmarks and improves visual understanding and everyday work tasks.
Anthropic has unveiled Claude Sonnet 5.5, its latest mid-tier artificial intelligence (AI) model, which boasts significant advancements in coding, knowledge work, and visual understanding. The new model is designed to be more efficient and cost-effective compared to its predecessor, Sonnet 5.
According to Anthropic, Claude Sonnet 5.5 generates responses more than 30% faster than its predecessor, making it a faster tool for various tasks. Additionally, the model can cost up to 30% less per task due to its reduced token usage, which allows it to complete the same work with fewer resources.
The company highlights that Claude Sonnet 5.5 has improved coding capabilities, scoring 70.6% on Terminal-Bench 4.0, compared to 10.3% for Sonnet 5. It also achieves 46.2% on FrontierCode 1.1 and 55.5% on CursorBench 4.0, showcasing its enhanced ability to handle coding tasks more efficiently.
Anthropic emphasizes that the new model can understand large codebases quickly and complete coding tasks in fewer steps by batching tool calls more efficiently. The company claims that Sonnet 5.5 can handle tens of thousands of lines of code and sustain multi-hour coding tasks while maintaining fast response times.
On the GDPval-AA benchmark, which evaluates real-world professional tasks across 44 occupations and nine industries, Sonnet 5.5 scores 1,844, nearly matching the performance of Claude Opus 5.5 at 1,846 and significantly outperforming Sonnet 5’s score of 1,449. The model also improves computer-use performance and visual chart recognition, with scores of 80.1% and 61.6%, respectively.
Beyond coding, Claude Sonnet 5.5 is positioned as a model for well-defined everyday tasks that require a balance of speed, capability, and cost. It is designed for tasks such as fixing bugs, creating documents, building presentations, and working with spreadsheets. The model also excels at following visual and design requirements, creating polished user interfaces and presentations that require less manual editing.
Anthropic notes that the new model demonstrates stronger long-horizon reasoning and image understanding. It is the first Sonnet model capable of beating Pokémon Red while working only from screenshots, highlighting its ability to interpret visual information and carry out tasks over extended sequences.
In terms of cost, although Sonnet 5.5 has the same listed token prices as Sonnet 5, Anthropic reports that the new model generally requires fewer tokens to complete tasks. This makes Sonnet 5.5 up to 30% cheaper per task than its predecessor. At lower effort settings, the model can also outperform Sonnet 5’s previous best results on some benchmarks at a fraction of the cost, further reducing the time required for routine tasks.
Stronger safeguards accompany the improved capabilities of Claude Sonnet 5.5. Anthropic states that the model’s cybersecurity abilities have significantly improved compared to Sonnet 5. As a result, Sonnet 5.5 is the first Sonnet model to launch with cybersecurity safeguards and fallbacks similar to those used for Anthropic’s more capable models. Routine software development and bug fixing remain supported, while higher-risk tasks are more carefully managed.
