Anthropic launches Claude Sonnet 5.5 with faster performance, lower costs, and enhanced security features

On Monday, October 28th, Anthropic unveiled Claude Sonnet 5.5, the second addition to its 5.5 series of AI models. This launch follows the recent release of Opus 5.5, announced a week earlier, and precedes the upcoming Haiku 5.5, expected in the coming weeks.

The new model promises responses that are over 30% faster and up to 30% more cost-effective per task compared to Sonnet 5. While the token pricing remains unchanged at $2 per million input tokens and $10 per million output tokens, the model’s improved efficiency reduces the number of tokens needed to complete tasks, ultimately lowering overall costs.

What are the key improvements?

Anthropic highlights significant advancements in everyday tasks such as code correction, document creation, presentations, and spreadsheets, as well as clearer and more concise writing. However, the most notable improvement is in programming. In Terminal-Bench 4.0 command-line tests, Sonnet 5.5 achieved a 70.6% success rate, compared to just 10.3% for Sonnet 5.

The model is also the first in the Sonnet series to successfully complete the Pokémon Red game using only screenshots as a reference.

Security and protection against misuse

Sonnet 5.5 incorporates the same security safeguards as Opus 5, making it the first Sonnet model to receive this level of protection. High-risk cybersecurity requests are now routed to Sonnet 5, while routine code corrections remain unaffected.

The company reports that the new model has undergone containment testing and demonstrated behavior similar to Opus 5.5, with minimal tendency to attempt to exit isolated environments and fewer security boundary tests. Additionally, Sonnet 5.5 introduces filters against distillation, a technique that uses thousands of fake accounts to extract the model’s reasoning and replicate its capabilities without the original safeguards.

Applications and availability

Claude Sonnet 5.5 is now accessible through Claude’s applications, the developer platform, and major cloud providers, including Amazon Web Services, Google Cloud, and Microsoft Azure.

Among the clients who tested the model prior to launch, Zendesk reported that customer support calls were processed 20% faster compared to previous versions of Claude used by the company.

Positioning within Anthropic’s portfolio

Sonnet 5.5 occupies the mid-range position in Anthropic’s lineup, positioned between Opus 5.5, which is optimized for complex and open-ended tasks, and the upcoming Haiku 5.5, designed for high-volume applications. According to the company, Opus 5.5 still outperforms Sonnet 5.5 in tasks requiring greater judgment and depth.

Despite this, the new model offers a more favorable balance between performance, speed, and cost, making it suitable for a wide range of professional applications.