AI for Mortals — Video Script

Claude Sonnet 5 Just Dropped — Is It Actually Worth It?

Cold-open hook + section outline · draft 1

01  The Hook (Cold Open, ~150 words)

So, Anthropic just dropped Sonnet 5, and within an hour my feed was already full of "this changes everything" posts. It always is. But here's the thing, Sonnet's always been the budget pick, the one you reach for when Opus is overkill and Haiku's too dumb. So when a "mid-tier" model launches with this much noise around it, my first question isn't is it good, it's is it actually worth switching for. I spent the day running it against Opus 4.8 and GPT 5.5 on the same real tasks, coding, writing, a couple office tasks too, same prompts, no do-overs, no cherry-picking. And the results genuinely surprised me, for better and for worse. So in this video, I'm going to show you exactly what's new in Sonnet 5, where it actually beats the bigger models, where it quietly falls apart, and whether you should switch today or just wait this one out.

~148 words Est. runtime: 55–65 sec

02  Section-by-Section Outline

  1. What Actually Shipped

    Quick rundown of the official release: context window, pricing, headline benchmark numbers — and which of those claims are marketing versus real.

  2. The Skeptic's Take

    The contrarian beat: benchmarks moved, but does that translate to anything you'd notice in a real workday? Set up the test plan to answer it.

  3. Round 1: Coding Head-to-Head

    Live, single-shot build with the same prompt fed to Sonnet 5, Opus 4.8, and GPT 5.5 — screen-recorded, no revisions, deployed live.

  4. Round 2: Real Office Work

    Same head-to-head but for non-coding tasks — a slide deck, a sales email, a financial memo — to see how it holds up outside a dev environment.

  5. Where It Quietly Falls Apart

    The honest part: a specific task or edge case where Sonnet 5 underperforms the bigger models, so you know exactly when not to reach for it.

  6. The Price-to-Performance Math

    Break down actual cost per task versus Opus and GPT 5.5 — when downgrading to Sonnet 5 saves real money and when it costs you more in rework.

  7. Should You Switch?

    The verdict, split by viewer type — vibe coder, office worker, power user — with a clear yes/no/wait recommendation for each.

  8. How to Actually Switch

    30-second walkthrough of swapping the model in Claude Code / claude.ai so viewers can try it themselves before the video ends.