Claude Opus 5. 5 in Make: Head-to-Head Test vs Sonnet & Opus 5 on LinkedIn Post Analysis

In this hands-on test, Vicente puts Claude Opus 5.5, the newest Anthropic model, to work inside a Make AI Agent. He runs the same task on three models, Sonnet, Opus 5, and Opus 5.5, and compares the quality of each report, the token usage, and the speed.

🔍 Inside the Build:

* The Scenario: A weekly Make scenario takes about 150 LinkedIn posts from Vicente’s favorite creators, each with its hook, taxonomy, and performance data. An AI agent analyzes them and posts a digest in Slack.
* The Goal of the Report: What’s working, what isn’t, and what Vicente could try next in his own content.
* Round 1, Sonnet: A baseline run on the last 50 days of posts, with a structured Slack report covering top creators, patterns, and recommendations.
* Round 2, Opus 5: See how it compares in speed, token consumption, and how specific the patterns are.
* Round 3, Opus 5.5: Vicente runs the newest model and looks at how thoroughly it reads the posts, how specific its patterns are, and how actionable its content templates are.
* Choosing Your Model: See where to switch models in the Make AI Agent module, and how to use your own API key or Make’s AI Toolkit if you can’t create one.

Vicente’s honest takeaway: for this fairly simple, qualitative use case, Opus 5.5 isn’t a huge game changer, but the output was more specific and needed less handholding. This video is a good starting point for anyone choosing a model for their agents, especially for tasks like triage, classification, and document or proposal generation that call for more judgment.

🛠️ Featured Stack:

* Make AI Agents (Model selection & comparison)
* Claude Opus 5.5, Opus 5 & Sonnet (Anthropic models)
* Make Scenario Designer (Weekly scheduled scenario)
* Slack (Delivering the weekly digest)

💡 Pro-Tip from Vicente: Give newer models a goal instead of detailed step-by-step instructions. In his tests, Opus 5.5 needed less handholding and worked out an efficient way to reach the goal on its own. Then test it in your more complex scenarios, where it can handle routing, triage, and other judgment calls. 🔗 Make


https://www.youtube.com/watch?v=l-u1WLOOTyg&utm_source=dlvr.it&utm_medium=blogger
Imágenes del tema de enot-poloskun. Con tecnología de Blogger.