Anthropic unveiled Claude Sonnet 5.5 on September 28.
The new model keeps the same token prices as the previous Sonnet 5 while focusing on handling tasks with fewer tokens and steps.
So what difference does the claim that it is faster and cheaper actually make to the cost of using it?
3-Line Summary
1. Claude Sonnet 5.5 is not a model with lower unit prices
2. The $2 input price and $10 output price are the same as those of its predecessor
3. For repetitive tasks, both cost and reasoning settings need to be considered
The Price Is the Same, but the Way It Completes the Same Task Has Changed
Claude Sonnet 5.5 is the second model in Anthropic’s Claude 5.5 product family. The company introduced it as a model designed for tasks with relatively clear boundaries, such as code modifications and creating documents, slides, and spreadsheets, rather than for complex, long-term judgment-based work.
The API price is $2 per 100 tokens,000 input tokens and $10 per 100 tokens,000 output tokens. Cache reads cost $0.20, and cache writes cost $2.50. These are the same unit prices as Sonnet 5. Therefore, “up to 30% cost savings” does not mean that the prices on the pricing table have fallen; rather, it reflects Anthropic’s explanation that the model uses fewer tokens and task steps for the same work.
The company also stated that output generation is more than 30% faster than Sonnet 5. However, these figures and the extent of the cost savings are based on Anthropic’s announcement and internal evaluations. Actual costs may vary depending on the length of requests, the number of repetitions, and the selected reasoning level.
The Scope of the Work Matters More Than Coding Benchmark Scores
Anthropic announced that Sonnet 5.5 scored 70.6% on TerminalBench 4.0, an agentic coding benchmark. This is higher than Sonnet 5’s 10.3%. On CursorBench 4.0, which involves working with multiple code files, it reported a score of 55.5%.
However, these scores alone do not show that it can replace a higher-end model for every task. Anthropic explained that Claude Opus 5.5 still has strengths in complex, open-ended tasks that require ongoing judgment. The core strength of Sonnet 5.5 is not taking on every highest-difficulty task, but rapidly repeating clearly defined work.
The effort-level setting, which adjusts the resources used for reasoning, is also connected to this distinction. Raising the setting can use more tokens and lengthen the processing. Even when the unit price is the same, repeatedly revising the result or continuously using a high reasoning level can make the cost per task differ from the savings range presented by the company.
Seoul Region Support Offers a Choice of Where Data Is Processed
AWS stated on September 30 that it would provide models, including Claude Opus 5 and Claude Sonnet 5, through Amazon Bedrock based in the AWS Asia Pacific (Seoul) Region. As a result, companies and institutions have the option of processing Claude model inference and customer data domestically.
This is particularly relevant to organizations such as those in finance, the public sector, and healthcare that need to examine requirements for domestic data processing. AWS explained that users only need to select a region-specific inference profile in Bedrock without making separate infrastructure changes. However, each organization must conduct its own security and regulatory review to determine which data and tasks can actually be covered.
When evaluating Claude Sonnet 5.5, it is more accurate to consider not only the simple claim that it is “cheaper,” but also whether the tasks to be repeated are clearly defined and how many tokens and reasoning steps those tasks require.
References
Tags #Claude #ClaudeSonnet55 #ClaudeSonnet55 #Anthropic #GenerativeAI #AIModel #AICoding #AgenticCoding #TokenCosts #APIPrice #AWSBedrock #SeoulRegion #DomesticDataProcessing