Modern Creator
Every · YouTube

GPT-6 Astra: 5 Things to Know on Day One

Every's Dan Shipper puts OpenAI's new flagship through a same-day vibe check and stacks it against Anthropic's Fable.

Posted
4 days ago
Duration
Format
Review
educational
Views
7K
316 likes
Part of the collectionThe GPT-6 Astra PlaybookEvery GPT-6 Astra breakdown, synthesized into one page.
Read the playbook
Big Idea

The argument in one line.

GPT-6 Astra is a genuinely strong daily-use model for writing, computer automation, and 3D visualization, but it over-decorates every interface it builds and drifts from a prompt's exact intent, which keeps it a step behind Fable for high-stakes, long-running delegation.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You already use Claude, GPT, or another frontier model daily and want to know whether GPT-6 Astra is worth switching to or adding to the rotation.
  • You build software, decks, or visual assets and want a concrete read on where a brand-new model's computer-use and design skills actually land.
  • You're deciding which model to hand a long, high-stakes, mostly-unsupervised task versus which one to use for quick daily work.
SKIP IF…
  • You're looking for a pricing table or spec sheet — this is a hands-on impressions video, not a benchmark writeup.
  • You don't currently use any frontier AI model and aren't shopping for one — the comparisons assume you already have Fable or something comparable.
TL;DR

The full version, fast.

Every's team spent release day testing OpenAI's new GPT-6 Astra and found it excels at writing, computer use, and 3D visualization: it can draft a passable self-review from Slack messages, edit a five-hour video in Premiere unsupervised, and build a historically accurate Battle of Waterloo scene from written accounts. Its weak spot is restraint. Astra decorates interfaces with extra buttons and text the prompt never asked for, and it sticks less closely to the literal intent of a request than Anthropic's Fable does on the same task. Verdict: an S-tier daily driver, but Fable still wins the biggest, longest-running delegation work.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →
Chapters

Where the time goes.

00:0001:13

01 · Intro + sponsor read

Announces GPT-6 Astra dropped hours earlier and previews the five-point vibe check, followed by a sponsor read and the Every subscription pitch before testing begins.

01:1302:13

02 · 1. Writing

Astra one-shots its own critical vibe check by reading the team's Slack messages and calls itself 'a show horse, not a workhorse'; praised as a crisp, slop-free daily writing companion.

02:1302:57

03 · 2. Computer use

Astra runs unattended in Adobe Premiere for about five hours, cutting together a video that reached 25,000 views on its own.

02:5703:32

04 · 3. 3D games and visualizations

Astra builds an explorable, historically accurate 3D reconstruction of the Battle of Waterloo in three to four hours from written accounts and geography.

03:3204:39

05 · 4. Interface design

Astra's visual taste is strong but it over-decorates interfaces with redundant labels, text, and buttons, especially at higher reasoning-effort settings, leaving Fable's first pass cleaner.

04:3906:19

06 · 5. Prompt adherence

Head-to-head, Astra and Fable each build a handwritten-journal digitizing app from the same brief; Fable's one-button, page-by-page flow beats Astra's more cluttered, multi-step interface.

06:1906:38

07 · Final verdict

Calls Astra an absolute S-tier daily driver at Fable's price point, but keeps Fable for the biggest, longest-running delegation work.

Atomic Insights

Lines worth screenshotting.

  • GPT-6 Astra wrote its own critical self-review by reading its team's Slack messages, and called itself 'a show horse, not a workhorse.'
  • Astra spent about five hours autonomously editing a video inside Adobe Premiere, producing a cut that reached 25,000 views.
  • Astra built an explorable, historically accurate 3D reconstruction of the Battle of Waterloo in three to four hours from written accounts and geography.
  • GPT-6 Astra sits in a new OpenAI model class above Sol, and is priced to compete directly with Anthropic's Fable.
  • Astra's interfaces get more cluttered with redundant labels and buttons at higher reasoning-effort settings, not fewer.
  • Asked to build the same handwritten-journal digitizing app, Fable produced a one-button, page-by-page workflow while Astra required more clicks and setup per page.
  • Every's reviewer rated GPT-6 Astra an S-tier daily driver but still defaults to Fable for the longest, highest-stakes delegation tasks.
  • Good visual taste and faithful adherence to a prompt's actual intent are separate skills, and a model can have one without the other.
Takeaway

Astra is a strong daily driver with one design tell

WHAT TO LEARN

Astra writes cleanly and can run a computer unattended for hours, but it can't resist decorating every interface it builds, which is the tell that separates a daily driver from a delegate-and-forget model.

021. Writing
  • A model that can draft its own self-assessment by reading a team's Slack messages is accurate enough to trust for first-pass writing, not just polish.
  • Crisp, slop-free prose is now a real differentiator between models, not just a marketing claim, so test a new model's writing on a task you already know well.
  • Letting a model grade its own output first surfaces its blind spots faster than waiting for a human editor to find them.
032. Computer use
  • A model that can run a five-hour video edit unattended turns a single license into a virtual employee, not just a faster keyboard shortcut.
  • Computer-use capability multiplies output on any task with a GUI, from slide decks to spreadsheets, not just coding, so reconsider what you assign to a person versus a model.
  • A 25,000-view video cut almost entirely by a model is a concrete benchmark for how far unattended computer use has come, not a demo-only party trick.
043. 3D games and visualizations
  • Historically accurate 3D reconstructions built from written accounts and geography data now take a few hours instead of a specialized studio team.
  • The line between proof-of-concept and usable output has moved: a model-built game world can be explored, not just viewed as a screenshot.
054. Interface design
  • A model with strong visual taste can still ship interfaces cluttered with redundant text, labels, and buttons; good design instincts and good UX restraint are separate skills.
  • The clutter gets worse at higher reasoning-effort settings, so more compute doesn't automatically mean a cleaner result, check output at the setting you'll actually use.
  • When comparing two models on the same brief, count the extra steps and buttons each one adds; that's a faster tell than eyeballing visual polish.
065. Prompt adherence
  • A model can have better design taste and still be harder to use, because taste and faithful adherence to what you actually asked for are different capabilities.
  • Test any model against the same real task on two tools side by side; the workflow gap (one button versus many) often matters more than surface polish.
  • A model that adds unrequested bells and whistles is quietly changing your spec, so restate the constraint explicitly if simplicity is the actual requirement.
07Final verdict
  • A model can be priced above the previous flagship tier and still be the right daily pick if it's S-tier for your everyday tasks.
  • Reserve the most expensive or highest-context model for long-running, high-stakes delegation, and use the daily driver for everything else; the two roles don't have to be the same tool.
Glossary

Terms worth knowing.

Vibe check
A hands-on, subjective first-impressions review of a new AI model built on real test tasks rather than benchmark scores.
Computer use
An AI model's ability to control a computer directly, clicking, typing, and navigating software the way a person would, to complete multi-step tasks unsupervised.
Reasoning effort
A setting that controls how much computation a model spends before responding, trading speed for depth on harder tasks.
Delegation task
A long, mostly-unsupervised piece of work handed entirely to an AI model to complete with minimal check-ins.
Resources

Things they pointed at.

00:19productCelsius
00:19channelEvery
00:48productCora (email agent)
01:01productEvery agent
02:25linkFable 5.1 vibe check video
Quotables

Lines you could clip.

01:18
GPT-6 Astra is a show horse. OpenAI's new model has striking visual taste and a knack for consulting work. Its ambitions still outrun its judgment.
Astra's own self-written verdict, delivered as a punchy label.TikTok hook↗ Tweet quote
01:56
I feel like its sentences are crisp. They're to the point. There's no AI-isms. There's no slop.
Tight praise line with an implied before/after contrast against typical AI writing.newsletter pull-quote↗ Tweet quote
02:25
It spent like five hours in Premiere earlier this week, cutting together the Fable 5.1 vibe check that we released earlier this week. That video has like 25,000 views.
Concrete, verifiable proof point for unattended computer use.IG reel cold open↗ Tweet quote
06:19
Overall verdict, absolute S tier daily driver... But for the top end, long running, big delegation tasks, I still use fable.
Clean closing verdict, works as a clip end-card.TikTok hook↗ Tweet quote
The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphor
it's model release day gpt 5 .6 astra the long -awaited astra is dropping today we just got word that it was dropping a few hours ago we've been tense testing it extensively so we're putting this together for you to have a review on day one found out at three o 'clock in the morning it was dropping so this vibe check brought to you by Celsius.
Here are five things that you need to know about Astra on day one. From our extensive testing, we've got a 30 -person team taking it through everything from writing to coding to design. And if you want more of this, you should go to Every, every .to.
It's the only subscription you need to stay at the edge of AI. If you want to be at the edge, you need education and you need equipment. Every provides both.
We've got a daily newsletter. We've got courses. We've got camps.
We've got everything that you need to know which models to use. for what on day one? And we've got equipment.
We've built our own tools from an email agent like Cora to the Every agent, an agentic coworker that lives in your Slack. It's all bundled together under one subscription. Check it out now at every .to.
And if you really want more, we have the full vibe check. It's a several thousand word article with all of our testing from our whole team live right now on the Every website, every .to. Now let's get into it.
Number one, it's a fantastic writer. It one -shotted its own vibe check. I got to show you this.
This is the vibe check that it wrote for itself based on looking at our Slack and deciding what we thought. Vibe check. GPT -6 Astra is a show horse.
OpenAI's new model has striking visual taste and a knack for consulting work. Its ambitions still outrun its judgment. It's a little harsh on itself, but this is good.
GPT -6 Astra is very good at getting carried away. Opening eyes new model. Out today builds beautiful things.
I asked for a 3D rendering of the Hadestown set and got a little theater with weathered green walls, hanging lamps, and a wooden stage ringed with lights. That's very good. That's very hard to do.
This model... I've been using it for a while, and I love it for writing. It is my constant companion.
I feel like its sentences are crisp. They're to the point. There's no AI -isms.
There's no slop. It's a very, very impressive writing model, and you need to check it out. All right, number two.
It is a beast at computer use. Truly, there's something special happening here with computer use. We've passed this point where it can basically use the computer for you.
A great example, it spent like five hours in Premiere earlier this week, cutting together the Fable 5 .1 vibe check that we released earlier this week. That video has like 25 ,000 views. It was cut mostly by Astra just running on our head of video, Randy's computer, and it had him going, whoa.
So for... Pretty much anything you're doing on your computer, from slide decks to video production to spreadsheets, this thing can do a lot for you and it multiplies what you're able to do in a day. All right, number three.
It is really good at 3D games and visualizations. I had it make me this reconstruction of the Battle of Waterloo that I can go around in and you can see the British, you can see the French, you can press play. It's all historically accurate from reading accounts from the battle and looking at the geography.
And this is crazy. This took, I don't know, three or four hours. And it's just wild that we live in a time where you can just make stuff like this.
incredibly good at 3D visualizations in games. Okay, next, number four. And now we're getting into some of the things that aren't quite as good about this model.
Just to take a step back, it's a new model class. It's above Sol and OpenAI's lineup. which means it competes with Anthropix Fable.
And we found in our testing that it has a few frustrating characteristics that, in our opinion, keep it from reaching quite Fable's top end. Its interfaces are a little bit extra. So a very simple example is I had it put together a website that had a login page.
And you can see this. it puts private case study and then enter shared password to continue and then password and then open the case and then no account or sign in or service provided like that maybe doesn't seem so egregious but it it happens a lot especially at higher effort levels and it puts this kind of text and labels and buttons in a lot of different places that don't make as much sense so what you're going to find is that fable's first take on things is just a little bit cleaner and more refined But Astra itself has good design taste.
It's just extra with all the bells and whistles it puts into interfaces. Number five is it's understanding and adherence to the underlying sense of your prompt and then sort of taking it beyond the prompt in a way that is delightful, surprising, and still simple. is just not as good as Fable.
It tends to add more bells and whistles and it tends to maybe slightly misunderstand the sense of your prompt. So I'll give you a specific example. Our editor Jack Chang asked both Astra and Fable to create an app for him to help him digitize his handwritten journals.
Here's the welcome screen from Fable. You can see it's like pretty simple. Here's the welcome stream from Astra.
You can see there's more design taste, but it's got all these buttons. It's got all this stuff. You got to type all this stuff.
It's like it's a little harder to use. But I think the more important thing is if you look at the process that Fable design versus Astra design, you can see what I'm saying about the intuitive adherence to the sense of the prompt and then extending it. What Fable Design, pretty simple workflow, just show it a page, press one button, and then go to the next page, and it'll just continuously transcribe.
Astra, it has a more well thought out, well designed. interface like it's got the warm paper it's got like the green highlights and all that kind of stuff but it sort of forces you to do one page at a time click a few buttons save it then go to the next one it's not built as intuitively for his use case and and that's just the thing we find with astro that keeps it from in our view taking the top spot in terms of this big powerhouse that's going to go for a long time and give you something really usable at the end So overall verdict, absolute S tier daily driver.
If you can afford it, remember this is a model class above soul. So it's priced like fable. If you can afford it, I use this model every single day for pretty much all my work.
But for the top end, long running, big delegation tasks, I still use fable.
The Hook

The bait, then the rug-pull.

OpenAI dropped GPT-6 Astra with almost no warning, and Every had a 30-person team testing it within hours. This is the same-day verdict: where Astra earns a spot in daily use, and where Anthropic's Fable still wins.

Frameworks

Named ideas worth stealing.

01:13list

5 Things to Know About GPT-6 Astra

  1. Writing
  2. Computer use
  3. 3D games and visualizations
  4. Interface design (weak point)
  5. Prompt adherence (weak point)

The video's own structure for a same-day model review: three clear strengths followed by two specific, demonstrated weaknesses versus a named competitor.

Steal forany same-day product or tool review that needs a fast, skimmable structure
CTA Breakdown

How they asked for the click.

VERBAL ASK
00:19product
check it out now at every.to

Soft-sell for the Every subscription (newsletter, courses, camps, in-house AI tools) folded into the cold open before the actual testing content begins, plus a pointer to the full-length written vibe check article.

MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Storyboard

Visual structure at a glance.

open
hookopen00:00
self-written vibe check
promiseself-written vibe check01:13
Premiere autopilot edit
valuePremiere autopilot edit02:25
final verdict
ctafinal verdict06:19
Frame Gallery

Visual moments.

One-click upgrade to your Google

Get more breakdowns in your search results

Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.

Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Watch next

More from this channel + related breakdowns.