Anthropic has released Claude Sonnet 5.5, the latest iteration of its mid-tier AI model. According to the company, the new version is designed to be more efficient, running over 30% faster than its predecessor while reducing operational costs by up to 30% for most tasks. Despite the performance gains and improved benchmark scores across the board, the model maintains the same pricing structure as the previous Sonnet 5.
Performance and Efficiency Gains
The release follows a trend in the industry toward optimizing model performance without increasing price points. By improving speed and cost-efficiency, Anthropic aims to make high-level reasoning more accessible for developers and enterprise users. Simon Willison reported that while the model generally outperforms Sonnet 5, it also introduces a thinking effort feature that allows users to adjust the amount of processing power applied to a task. In tests involving the generation of SVG graphics, such as a pelican on a bicycle, the model demonstrated varying levels of success based on these settings. An xhigh effort setting successfully generated an image in 41 seconds at a cost of 5.74 cents.
Technical Limitations and Missing Details
The launch has not been without technical hurdles. Testing revealed a bug similar to one found in Opus 5.5, where the max thinking effort setting resulted in a failure. In one instance, the model processed 128,000 tokens over a long period, costing $1.28, before hitting its token limit and failing to produce the requested SVG. It remains unclear if this issue is specific to complex graphical tasks or if it affects other high-reasoning workloads. Furthermore, the announcement did not specify the exact technical architecture changes that led to the 30% speed increase. Anthropic has not yet provided a timeline for a patch to address the token exhaustion bug in the maximum thinking mode.
As users begin integrating Sonnet 5.5 into their workflows, the focus will likely remain on whether the promised cost savings hold up across diverse use cases. The model’s ability to balance speed with deep reasoning will be a key metric for its adoption.
Lead image: Brett Sayles / Pexels
