OpenAI Dev Day in 4 Minutes

--- type: transcript date: 2025-10-06 youtube_id: g3HEvM0qB48 --- # Transcript: OpenAI Dev Day in 4 Minutes In this video, I'm going to do a quick recap of everything that OpenAI announced during their dev day. First up, in terms of one of their product announcements was apps within Chat GPT. So, what apps within Chat GPT is is effectively a way that you can access different sites as well as their functionality directly within Chat GBT. Say for instance, if you want to connect to something like Spotify and make something like a playlist. So chatbd can do really well at creating things like playlist, but this takes it a step further where it will actually go and create a functional playlist on Spotify. You can use this for something like say Zillow and find different information about different homes in various cities. And there's a ton of different use cases for this whether it's using Canva within Chai GPT leveraging even expedia or a number of other sites that they had included within the announcement as well. Next up, arguably their biggest announcement was Agent Kit. Agent Kit is a really neat platform. It's very similar to something like NADN where it effectively allows you to build AI agents without having to write any code. You still sort of have to think through some of the conditional statements and the logic of the overall agent orchestration, but it makes a really easy interface for actually setting up all of the different features. You're going to be able to access things like file search, different MCPs that you can tie in within this. You can also create custom agents along different conditional routes of the architecture that you want to set up as well. Next up, you can generate videos from Sora 2 from the API. So, this is going to unlock a ton of different applications where you can customize these different models as well as their prompt depending on your use case. So, there is Sora 2 as well as Sora 2 Pro. It is as simple as selecting the model as well as passing in the prompt. And then once you submit the request, you'll be able to pull for the status of the generation. Now, obviously, this does take a little bit of time to generate the video, so just be mindful of that. Next up is Codeex. There are a ton of different updates within Codex, mostly around how you can actually access the agent. You can now access it within Slack. And in addition to that, there is a codeex SDK. If you do want to create your own custom agents where say if you want to build something like your own version of something like lovable or one of these app builders, you can leverage the codeex SDK for something like that. Now, additionally, there is also usage. So, you're going to be able to see all of the different analytics of actually your usage within codecs as well. Another one that I think a lot of people will be excited about is GPT5 Pro is now within the API. Now, the one thing that I do want to note with this, this is a very expensive option. So, $15 per million tokens of input and $120 per million tokens of output. and you are able to access up to 400,000 tokens of context to pass into the model and be able to generate up to 272,000. Now, at time of recording, I haven't seen any independent benchmarks of the model and how it stacks up to some of the competition, but I would not be surprised if this is the best model in the world, basically across benchmarks. There could be some exceptions, but I think this definitely stands quite a good chance to rank right at the top. Next up, they also launched GPT Realtime Mini, which is a 70% cheaper version of their real-time API. If you are building voice applications, this is definitely a very competitive offering where you're going to be able to have all of those real life intonations, be able to understand tone, and all of those different intricacies when you're building out a voice application. Okay. And next up, there is GBT Image 1 Mini. Again, this is a much cheaper version of GBT image 1- which is their famous image generation model which allows you to create anything from infographics all the way through to very hyperrealistic images. Similar to the real-time voice API, this is a cheaper version of its predecessor. Okay, so last I want to do a really quick demonstration of what the agent builder looks like. Here's a demonstration of it using a conditional statement, it routing to the particular agent. And then from here, the really neat thing that I like with this is you can also define these different widgets. So say if you want conditionally rendered UI elements, like say in this case, you want to actually show a different format for the different flights. You can go ahead and define that and those relevant classes. And you can actually edit the instructions in terms of how to actually render those different widgets. Within here, you can see a demonstration of that where if you ask and meet that criteria, what it will do is it will pass in those relevant arguments to actually pass this really nice card that you can define as a part of the widget building process. That's pretty much it for this video. If you found this video useful, please like, comment, share, and subscribe. Otherwise, until the next