GPT-5.6
Computer programWikipedia
843
MENTIONS
440
EPISODES
234
PODCASTS
Search complete. 843 mentions across 440 episodes found for "GPT-5.6".
Sep 11, 2026
David GerardHOST
Oh, GPT-6 is the AI now, you know? And, um, I know people are theorizing that they did that because of their deal with Microsoft, where they're out of their Microsoft deal once, um, they have AGI, and if they can get their totally independent mate committee to say that it's AGI, then, uh, it'll be, uh, fine.
David GerardHOST
And it's definitely the AGI, even though it's literally GPT 5.6 with a different label and marketing label.
David GerardHOST
It's, it's dumb as hell.
David GerardHOST
It's all dumb as hell all the way down, and the bits along the way are dumb too.
Jon KrohnHOST
GPT-6 Astra is the first six series model from OpenAI.
Jon KrohnHOST
It's OpenAI's most capable model then, of course, and it's positioned as the successor to GPT 5.6 Sol, which had been the company's flagship since July.
Jon KrohnHOST
And also, the name seems to imply to me that Astra is even bigger in terms of parameter count than Sol, which is bigger than Terra, which is bigger than Luna, so Moon, Earth, Sun, stars.
Jon KrohnHOST
I don't know.
Alex VolkovHOST
Folks, Terminal Bench, 2.1 score, 90.6.
Alex VolkovHOST
This model beats Opus 5 and GPT 5.6.0.
Yam PelegHOST
You need to see it on a table.
Yam PelegHOST
I have a table to show you.
9 MINS LATER
Alex VolkovHOST
The numbers kind of speak for themselves as well.
Alex VolkovHOST
Terminal bench 3.0, deep sick V4.1 flash gets 30%.
Alex VolkovHOST
That's beating Kimi K3 and that catches up to GPT 5.6 Sol at 34%.
Alex VolkovHOST
74.2 score on DeepSwe 1.1. DeepSwe notoriously a benchmark in eval that represents how actual coders feel.
Dan SanchezHOST
You notice when I cut over to work, you can actually select the exact type of model you want, right? You can go over to Astra and all the way down to ChatGPT 5.5.
Dan SanchezHOST
And of course, the three different versions of chat GPT 5.6 of Luna, Tara and soul and their intensity level you can take, you know, you can pick Astra and go all the way up to, like max ultra or what they call ultra, you know, reasoning, you know, if you need to go all the way.
Dan SanchezHOST
So work gives you more flexibility.
Dan SanchezHOST
But there's one thing that you need to understand the difference between chat and work.
NeilHOST
Mm.
AlexHOST
It's not a direct hardware comparison to something like GPT 5.6 Sol.
NeilHOST
Makes sense.
AlexHOST
But the model's actual real-world performance is a completely different story.
AlexHOST
Based on DeepSeek's own evaluation metrics, V4.1 Flash is remarkably capable.
NeilHOST
Mm.
AlexHOST
It actually beats GPT 5.6 Sol on several critical agentic benchmarks.
NeilHOST
Let's break those benchmark victories down because they reveal what the model is actually good at.
speaker_0HOST
But let's talk about how smart it actually is, because that's the crazy part.
speaker_0HOST
It just edged out massive, expensive frontier models like Cloud Opus 5 and GPT 5.6, Sol on agentic coding and cybersecurity benchmarks.
speaker_1HOST
Which is incredible given its size.
speaker_0HOST
Right.
MattiasHOST
Mm-hmm
AndreHOST
... uh, GPT 5.6, which like in my experience, like people saying it's kind of comparable to Fable.
AndreHOST
That's like not my experience really.
AndreHOST
Um, anyhow, um, there are no big reports of breaches, you know, like people like being hacked left and right.
Ian BrownHOST
So terminal bench gives Fable 5.1 55.8 and Mythos 5.1 60.9.
Ian BrownHOST
Whereas the Kershler bench gives it 73.4. And GPT 5.6 sole gets 67.2 instead of 37.3 on the terminal bench.
Ian BrownHOST
So they are way different numbers.
Stephen BancroftHOST
I don't know what the difference is in the
Stephen BancroftHOST
Terminal Bench Science, for instance, it's scoring, what's that, about 65% and it's doing it a little bit cheaper than Claude Fable 5.1. Have a look
Ian BrownHOST
at the Arc AGI 3 benchmarks.
Ian BrownHOST
Have a look at GPT 5.6 Sol is scoring 7.8 on the Arc AGI 3 test, and Astra is scoring 99.9. Wow.
Ian BrownHOST
Yeah.
Sam BownePANELIST
So you can't contain them with a virtual machine, which is what I've been using, unless you use a special virtual machine called Firecracker that is stripped down to present a smaller attack service.
Sam BownePANELIST
And Firecracker, even GPT 5.6 cyber, cannot escape.
Sam BownePANELIST
How are they
Paul AsadoorianHOST
escaping?
Type Three AudioNARRATOR
Astra improves significantly as you increase the number of filler tokens up to 4,096.
Type Three AudioNARRATOR
We also compare Astra with 4 Hop Natural Facts to Opus 4.5, Opus 5, GPT 5.6 Sol, and DeepSeek the V3.2 with 2 Hop Natural Facts.
Type Three AudioNARRATOR
There's an image here.
Type Three AudioNARRATOR
Line graph and hop natural facts.
Type Three AudioNARRATOR
See the original text for the table content.
Type Three AudioNARRATOR
Take away.
Type Three AudioNARRATOR
Besides USTRA, only GPT 5.6 SOL shows improvement on three of the benchmarks we test.
Type Three AudioNARRATOR
SOL does not show nearly as much improvement as USTRA.
430 more episodes mention GPT-5.6.
Create an account to see the whole feed, search across every transcript, and follow the entities you care about.