In a recent LinkedIn post, Federico Donatone delves into the strategic implications behind Anthropic’s significant price reduction for its Claude 3.5 Sonnet model. Donatone breaks down the economics of AI language models, emphasizing how cost structures directly influence market competitiveness.
Understanding AI Pricing Models
Federico Donatone highlights a fundamental aspect of AI cost: the per-word billing for both input and output. He points out the discrepancy between Anthropic’s stated cost reduction and the actual savings users might see, attributing it to model efficiency.
“Companies pay AI by the word: every word it reads and writes goes on the bill.”
As Federico Donatone explains, while the per-word cost for Opus 5.5 has decreased by 20%, the overall cost for a completed task is down by 40%. This suggests that efficiency gains beyond simple word count are driving the larger savings. Donatone attributes this to the model’s ability to be more concise.
The Impact of ‘Memory’ on AI Costs
A key area of analysis by Federico Donatone concerns the concept of ‘memory’ in AI assistants. He explains that AI models often need to re-process instructions, and each re-read incurs a cost. This ‘memory tax’ can significantly inflate operational expenses.
“AI assistants that work on their own reread the same instructions again and again, and every reread gets billed.”
Federico Donatone notes that Anthropic has identified this ‘memory’ cost as a substantial portion of their billing. The company’s claim of a 60% price cut in this area is, according to Donatone, a major factor in the overall cost reduction strategy for Claude 3.5 Sonnet.
Efficiency and Market Strategy
Beyond conciseness and memory management, Donatone also touches upon the hardware and speed efficiencies of the new model. He points out that the model requires fewer computational resources to run and delivers responses 30% faster.
Chip Availability and Competitive Edge
Federico Donatone connects Anthropic’s strategic timing of the release to market dynamics, specifically mentioning the acquisition of a large supply of Amazon chips. This suggests a proactive approach to securing resources necessary for scaling and maintaining competitive pricing.
“A cheaper model wins the price war. A shorter one wins it twice.”
In Donatone’s view, the combination of reduced word usage, efficient memory handling, and faster processing, bolstered by a strategic chip supply, positions Claude 3.5 Sonnet as a formidable competitor. He concludes by posing a question about whether this release represents the year’s best model launch, inviting further discussion on its market impact.
📝 About This Content
This article is based on insights shared by Federico Donatone on LinkedIn.
📅 Originally posted on October 1, 2026 | View original post on LinkedIn →