“It Blew Me Away” | Box’s First Look at GPT-6 Astra

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: “It Blew Me Away” | Box’s First Look at GPT-6 Astra · Channel: OpenAI · 02:07 · Watch on YouTube

Topics: Enterprise AI applications, AI evaluation and benchmarks, AI models, RAG, search and information retrieval

Summary

Yash, who leads AI at Box, describes Astra as a major advance for demanding customer workloads involving valuable, complex content. He demonstrates a financial analysis task combining profit, cost, and incentive-tax files, where the model distinguished between an incentive already applied in Argentina and one that still needed to be applied in Brazil. He says that getting this nuanced calculation right is one of the first times he has seen a model succeed at the task. Yash is excited to offer Astra to Box customers with needs for high judgment, nuanced reasoning, and extreme accuracy.

Key points

Yash leads AI at Box and says Astra’s capabilities impressed him.
Box customers work with valuable, intense, and complex content.
The example used profit analysis, cost analysis, and incentive-tax structure files.
The task required distinguishing an incentive already applied from one that still needed to be applied.
Astra retained the 2025 costs in Argentina because the 20% incentive was already applied.
Astra applied the 30% incentive in Brazil.
Applying the Argentina incentive twice would have caused the total cost to be wrong.
Yash says this was one of the first times he had seen a model get this task right.
He is excited to offer Astra to Box customers who need high judgment, nuanced thinking, and extreme accuracy.

Chapters

Start	Chapter	Summary
00:00	First impressions of Astra	Yash introduces Astra as an impressive model for the complex workloads Box customers face.
00:34	A complex financial analysis example	Yash explains how the model analyzed profit, cost, and tax-incentive files with different application statuses.
01:05	Getting the incentive calculations right	Astra correctly accounted for the incentive already applied in Argentina and the one to apply in Brazil.
01:36	Deploying Astra for Box customers	Yash says the result is one of the first he has seen for this task and shares his excitement about making Astra available to customers.

Quotes

Absolutely at the frontier. — 00:00

It blew me away — 00:34

The 20% incentive is already applied. — 01:05

This is one of the first times. — 01:36

Concepts and entities

GPT-6 Astra, Box, OpenAI, Yash, incentivos fiscais

Full transcript (original language)

00:00 Absolutely at the frontier. Just a solid model for some of these use cases [music] that have been just unbreakable in our stack. >> So, Yash, you lead AI at Box. [music] What were your first impressions about Astra? >> It blew me away in terms of just >> [music] >> how far this model has come in just doing these really heavy workloads that we see for our customers, [music] right? Some of our customers have content that

00:34 is very valuable, very intense, very complex. [music] >> I love that. I’d love to see an example of one. >> We gave it a couple files. We said, “Here’s some profit analysis. Here’s some cost analysis.” Then I give them a file that says, “Here is our some incentive tax structures.” [music] Now, the really interesting trick was there was two incentive tax structures. One of them was not applied to the profits. One of them was already applied. >> Mhm. >> Right? And it’s a such a easy thing to like miss. >> [music] >> But here’s where this is really frontier

01:05 status, and I haven’t seen this before, is that it actually said, “Hey, but in Argentina I’m retaining the 2025 costs because it’s already applied. The 20% incentive is already applied. Whereas in Brazil, I’m >> [music] >> I’m going to apply the 30% incentive.” And so, imagine if it did that wrong, right? It would have double-counted this 20% [music] incentive and gotten your entire cost wrong. >> Right. And I know this is a pretty complex like [music] prompt because of those calculations, but is it the very

01:36 first time you’re seeing a model, in this case Astra, get this right? >> Yes. This is one of the first times. [music] And that’s why I’m really, really excited about it. >> So, how are you thinking about deploying Astra to the Box customers now? >> I’m super excited because previously [music] I had customers, you know, grapple with these kinds of use cases where there was like a need for high judgment, nuanced thinking, and [music] extreme accuracy. And now I can go back to them and say, “Hey, you know, try this model and with our with [music] your content and you know, it should work.” And I I’m super excited. So,

02:07 yeah. >> Incredible. Well, we love the partnership with Box, and we can’t wait to see what you build. >> We, too. >> Thank you, Yash. >> Thank you.