In a recent LinkedIn post, Federico Donatone explores the often-overlooked economics of AI models, specifically comparing Opus 5 and Fable 5. Donatone, who recently conducted a three-day trial of Opus 5, challenges the conventional wisdom that a lower per-token cost always translates to greater overall affordability.
He highlights a critical distinction in AI output and session dynamics. According to Donatone’s analysis, the perceived cost-effectiveness can be misleading when not viewed in the context of actual work performed and output generated.
“Opus 5 should cost half of Fable 5, but there’s a catch.”
Donatone’s investigation delves into the quantitative differences between the two models over a typical session. He meticulously breaks down the metrics, revealing that while Opus 5 might appear cheaper on a per-token basis, its higher output volume and engagement requirements lead to a different economic reality.
Analyzing AI Output and Session Length
Federico Donatone’s findings suggest that the number of turns and the length of each message significantly impact the overall value derived from an AI session. His data points to a substantial difference in the amount of work and information processed.
As Donatone notes, his testing revealed:
- It takes 127 turns per session for Opus 5, compared to 67 for Fable 5.
- Each message in Opus 5 runs 29% longer than those in Fable 5.
This increased engagement and output volume are central to his argument about cost. Donatone points out that the higher number of turns and longer messages mean users are effectively getting more done within a single session.
“That is 2.46x the output for the same work.”
He quantifies this by stating, “It writes me 15,728 characters per session. Fable writes 7,726.” This means users are engaging with and processing more than double the content when using Opus 5, even if the per-token cost is higher.
The True Cost of AI: Beyond Per-Token Metrics
The core lesson Federico Donatone imparts from his experience is a warning against a simplistic view of AI costs. He argues that focusing solely on the price per token can lead to suboptimal decisions.
In his view, the true cost is determined by the value and volume of output relative to the effort and resources invested. The increased output from Opus 5, despite its potentially higher per-token price, offers a greater return on investment in terms of work accomplished.
“Cheaper per token is not cheaper.”
Donatone’s setup for the week involved GPT 5.6 orchestrating, Opus 5 on trial, and Fable building. This context underscores his practical approach to evaluating AI tools, moving beyond theoretical pricing to real-world performance.
His post concludes with an open question to his network, inviting discussion on their experiences with Opus 5, demonstrating a collaborative approach to understanding the evolving landscape of AI technology and its economic implications.
📝 About This Content
This article is based on insights shared by Federico Donatone on LinkedIn.
📅 Originally posted on July 27, 2026 | View original post on LinkedIn →