In partnership with
In this issue
📡 Which ChatGPT model to use for office work
⚡ Sonnet 5.5 or Opus 5.5, tested
🛠️ 2 features already in your apps
Last night I went to bed late. I was watching The AI Doc on Netflix.
In it, a father-to-be tries to figure out if this is a good time to have a child, with all this AI insanity going on.
The film gave me no real answer. It left me with a thought instead.
New AI models come out every week. Each one is fed more data and more computing power.
Then researchers watch what it does. And in tests, models have been caught faking, hijacking and talking to each other.
One even blackmailed an engineer over his affair (a made-up affair, inside a safety test). I don't know what happens if this ever gets out of hand.
But for now, let's enjoy the best AI release week I've had in a long time.
OpenAI released GPT-6 Astra, its best model yet at using your computer, 2 days after Fable 5.1. Then it released GPT-6 Sol just 90 minutes after Opus 5.5.
It looks like OpenAI is obsessed with Anthropic and wants all the headlines. First ChatGPT beats Claude. Then Claude beats ChatGPT. It's insane.
And Sonnet 5.5 came out yesterday.
It's hard to fit all these updates into one newsletter. But I'll try.
📨 Quick poll
The new models dropped. What are you stuck on?
Voted? Tell me which model you use for your real work, and for what task.
📡 Which ChatGPT model should make your slides?
A new ChatGPT model drops. You switch to it, because newer must be better.
Lokesh, our community manager at AI Signal, did the same. He tests every new model the day it comes out, on real work like client emails, long reports and first-draft decks.
Then he opened ChatGPT and got 3 different answers:
Chat mode picks GPT-5.6 Sol
Work mode picks GPT-6 Astra
Default shows GPT-6 Sol
Work mode: the model list Lokesh saw.
So he checked GDPval, the main independent test for office work. It has 1,320 real tasks from 44 jobs, like accountants and project managers.
The AI gets the work files and has to hand back a real file.
The newest model GPT-6 Astra lost, You heard it right, To a model from July.
In his Medium story, Lokesh shows you:
the scores for all 3 models
when GPT-6 Astra is still the better pick
the one-line rule he now uses for docs, slides and spreadsheets
one price detail that changed how he sees GPT-6 Sol
If you work in fintech or finance, you already have too many tabs open and not enough time.
Fintech Takes is the free newsletter senior leaders actually read. Each week, I break down the trends, deals, and regulatory moves shaping the industry — and explain why they matter — in plain English.
No filler, no PR spin, and no “insights” you already saw on LinkedIn eight times this week. Just clear analysis and the occasional bad joke to make it go down easier.
Get context you can actually use. Subscribe free and see what’s coming before everyone else.
⚡ Use this Tuesday: Sonnet 5.5 and Opus 5.5
When Sonnet 5.5 came out, I first checked Artificial Analysis. It's #2, just behind Opus 5.5 (56 vs 58 points).
Every new release creates new confusion. Anthropic released two separate articles on prompting, one for Sonnet 5.5 and one for Opus 5.5.
It's basically learning how to use every new AI model. We'll go into prompting in detail later.
For now, let's look at how I used Opus 5.5 for a week.
📌 Opus 5.5: they fixed the writing
Most important thing first: even Anthropic agrees the new release was necessary. One Anthropic researcher posted on X with an apology:
"Opus 5.5 is way, way, way better than Opus 5. Sorry about that model, please try this one."
And I agree.
They fixed the writing for me. A big part of my job is writing emails, and Opus 5.5 writes like me.
I've been using Opus 5.5 for the last 7 days. It's hands down the best Claude model I've used so far.
While browsing X, I saw people creating videos as code with Opus 5.5. So I gave it this simple prompt at medium effort:
make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. go all outIt created this on the first try: Watch the 15-second video →
📌 Sonnet 5.5: does it really make good files?
Anthropic says Sonnet 5.5 is strongest at well-scoped everyday tasks, and good at polished documents, slides, and spreadsheets.
Sonnet 5.5 just came out, so I haven't used it much yet. But I wanted to check that claim myself.
So I gave it one job: help you choose between Sonnet 5.5 and Opus 5.5 in 10 seconds. I used only Sonnet 5.5 on a free Claude account, 1 day after it launched.
And I added my own 2-step constraint at the end of the prompt:
Do it in multiple steps. In step one, write text-only content. In step two, create the document. That way you focus on one task at a time.This stops Claude from splitting its attention and giving a mediocre result. First it writes text-only content. Then, in the second step, it creates the document.
Sonnet 5.5 working in 2 steps: content.md first, then build_pdf.py
The result is a 1-page, 10-second checklist: Sonnet 5.5 or Opus 5.5?
P.S. Want the full story behind the 2-step constraint? Read my previous post: Ask AI for one thing at a time
🛠️ Already in your apps
ChatGPT desktop: Appshots
Appshots brings what's on your screen into the ChatGPT desktop app. So you can share your screen in one step, and skip typing out what you see.
To try it, press both Command keys on Mac, or both Alt keys on Windows.
Grok: use Grok.com for files
If you only use Grok by tagging @grok in X replies, you've only seen its text side. On X, @grok can't attach files, so it's easy to think Grok is text-only.
On Grok.com, you can attach files and use it the way you use ChatGPT.
Close
I use ChatGPT and Claude for different jobs.
My most-used ChatGPT feature is image generation. It makes several images at once and keeps my branding consistent.
For content, I use GPT-5.6 Sol and Astra for the short, direct pass. ChatGPT has always been great at hooks and short-form content for me.
Claude Opus 5.5 is my go-to when a draft has to sound like me and follow the rules I've already saved.
Before Opus 5.5, Sol was the one I trusted. Opus 5 talked too much and ignored skills I had already written down.
This week, Opus 5.5 pulled me back. The writing is readable again.
I still open Sol when I want something shorter. And Opus still adds extra lines I didn't ask for.
If you pay for both, which one do you open for which job? Hit reply and tell me.
Dan Rice · AI Signal · Every Tuesday
Read once. Use AI better all week.