Episode 50:
Opus, Sonnet and Haiku 5.5

Hey friends,

Last week on launch day, I typed one prompt into Claude Opus 5.5 and went to make a coffee ☕ (true story).

When I came back, it had made a full launch video. Animation, a little 8-bit Clawd, even video game music. ONE prompt on high effort. No skills, no MCP.

I posted it on Threads expecting a few likes from you lovely lot.

​It's now sitting at over 111K views and 1K+ likes. I still can't believe it LOL.

The number I kept staring at was a smaller one though. 12%. ONLY.

That's how much of my 5-hour limit the whole video used.

So I did what any normal person would do. I gave the exact same prompt to every new model that came out after it.

And Claude kept shipping. Three new models in 15 days.

Let me show you what that one prompt told me about each of them, and why Opus is still the one I'm most stoked about.

📌 The 30-Second Catch-Up

• ​Claude Opus 5.5 (Sept 22) - Works at Fable 5.1 level and costs 40% less than Opus 5
• ​Claude Sonnet 5.5(Sept 28) - Same price as Sonnet 5, WAY better
• ​Claude Haiku 5.5(Oct 7) - 75% cheaper than Haiku 4.5, pick the effort level
• Bigger limits - Paid plans got higher 5-hour limits, plus a reset you can use until Oct 22

✨ What One Prompt Told Me About Each Model

✅ 1. Opus 5.5: The Big Boy (for now)

The launch video was my first hour with Opus 5.5. I wanted to see if it was really as good as the hype.

That same night I gave GPT 6 Sol (on High, in Codex) the exact same prompt. Same rules. One prompt, no skills, no MCP.

It was nowhere near as good. Not even close.

About a week later GPT 6.1 Sol came out, so I had Opus make ITS launch video too. Opus won again.

Two weeks in, here's what I've noticed using it every day:

• Big model smell: it finishes long jobs and checks its own work like REALLY well
• Great to talk to: it chats like a smart friend and writes like a person
• Amazing at design and animation: my launch videos are just a web page it builds and records

Fair warning, it's addictive. My laptop barely gets a break anymore 😅

Pro tip: My sweet spot is High effort. And before you save anything as a skill, ask Opus this first:

Pick from its list, then save the skill. It works so well with Opus 5.5.

✅ 2. Sonnet 5.5: A Cheaper Opus?

When Sonnet 5.5 dropped six days later, my first thought was "okay, BRB" and I ran the viral prompt again.

I used extra effort this time, and the result made me say out loud: Anthropic is clearly feeding some crazy designs to these models.

It's very good, at least for a Sonnet. Way better than old Sonnet 5.

It still isn't Opus. When the job is big and messy, you can feel the difference. Anthropic says the same thing on their own launch page.

Pro tip: Let Sonnet do your everyday jobs and save Opus for the big ones. And nobody needs Sonnet on Ultracode. High and xhigh are plenty.

✅ 3. Haiku 5.5: The Tiny One That Surprised Me

Haiku 5.5 came out earlier this week, so of course it got the prompt too. This time I asked it to make a launch video about Haiku 5.5 itself. The whole video took 3% of my usage. 3% ONLY! 🤯

The output was decent for such a small model.

Then I pushed it on Ultracode for a GPT 6 in ChatGPT launch video. It only used 1%.

Anthropic is pitching it for fast stuff like summaries and computer use. It's also built to work as a helper (a subagent) next to Opus or Sonnet.

Pro tip: Don't ask Haiku to be Opus. Give it the small, repeat tasks inside a bigger task and let Opus RUN the show.

🛠 Make Your Own One-Prompt Test (3 Steps)

I never planned for that video to become my test. It just turned out to be the fastest way to know if a new model is worth switching to.

You can have one too:

1. Pick the job you've been putting off: The one that felt too big to try
2. Save one prompt: Don't tweak it between models
3. Run it on every new model: Same effort level, then compare the results and the usage

Here's my full prompt:

Swap "Opus 5.5" for whichever model you're testing.

📈 Why This Matters

A year ago, a video like mine meant hiring someone or learning After Effects.

Now I make one from my desk with a cold coffee, and the bill is a slice of my 5 hour limit.

The models also keep getting cheaper while they get better. Opus 5.5 works at Fable 5.1 level and still costs less to run than Opus 5. Haiku does a decent job for 3%.

12%. Then 3%. Then 1%. That's the real story of these launches. Trying something big used to cost money or a weekend. Now it costs less than a coffee break. So the question now is what you'd make if trying was basically free.

If the leaks are right, Fable 5.5 is next. It's getting the prompt the day it lands, and you'll see what it makes in the next episode 👀

🫡 Final Thoughts

That coffee I went to make on launch day turned into the best two weeks I've had on Threads.

Thank you to everyone who shared and replied to that video. And if you found this newsletter because of it, welcome. I'm really glad you're here.

Opus 5.5 is the one I'm most stoked about. It's the first model in a while that makes me want to open my laptop just to see what it'll make next.

Try the prompt this week on whichever model you've got. Then reply with what it made and which effort level you used. I'll share my favourites in a future episode.

And if you've got a friend who still thinks AI can't make anything good, forward this to them. They'll thank you 😉

To one good prompt and a lot of new models,
Nahid

​​​​​

600 1st Ave, Ste 330 PMB 92768, Seattle, WA 98104-2246
​Unsubscribe · Preferences​