Summary
Nate B Jones puts Anthropic’s Opus 5.5 through its paces and argues that the most important measure of a model is how many tokens it needs to finish a job, not just its price per million. The centerpiece is a demo in which Opus 5.5 turned a beanie-hat logo into a 514-piece Lego model, complete with a 63-page instruction booklet, a parts list, a BrickLink wanted list, and geometry-checking files.
The video breaks down the economics. The build consumed roughly 89 million tokens, about 1% of a weekly Claude subscription allowance, versus an estimated $44 or more at API rates. Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens. That is 20% below Opus 5 and 60% below Fable 5.1’s $10 and $50 rates. Jones also discusses Anthropic’s claim that typical workloads cost about 40% less, and explains why that figure does not simply mean 40% fewer tokens.
Viewers also get a practical framework for testing efficiency on their own tasks by comparing token usage across input, output, cache, and context buckets. Other topics include the model’s steerability in writing, how coding ability enables precise visual revisions, and how Opus 5.5 handles long unattended overnight jobs.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published September 30, 2026
🏷️ Format: Review
