The Ledger, every show

“But I do think that's going to be the killer use case will be like this generative UI experience that doesn't exist today because the models just take too long to generate tokens.”

“With reasoning is a great example of this where like multimodal with video understanding ended up like having this huge, like it's having this beautiful moment. The model is like soda out of the box because of all the reasoning capabilities that were baked in…”

“If you look at like multimodal, like the fact that the models can like with better, better than just from a multimodal input perspective, better than humans are at like most vision tasks, like The number of products and like things that that unlocks is like tr…”

“But I think if you look at like a, an example of this in practice, like 2.5 pro is actually an example where it wasn't just like RL scaling that made that model better. Yes, RL was part of the story, but like there was also a bunch of pre-training innovation a…”

“I don't think those people are going to come in through, you know, some front end website somewhere. Like those people are going to come in through audio from a telephone to texting to email. Like that's just the most obvious outcome.”

“Our models are probably never going to be as capable as some of the things that Harvey's doing because like our goal and our mission is really to solve this like very general use case.”

“I talked to a lot of developers who are building like, you know, just general purpose assistance and like general purpose agents and stuff like that. And I think it's cool and it's a good idea. I think like the challenge for them is like, they are going to end…”

“The actual model is two dollars per million input tokens and 12 dollars per million output tokens. So it's priced below some of those sort of comparable competitive models like GBT, 5.1 pro as an example, and Claude 4.5.”

“The model doesn't like the image quality doesn't get worse as you like do a multi-turn edit. So like you don't need to get everything right on the first edit.”

“Google controls. From a product perspective, like all the way to how the models are delivered to how the models are trained down to the silicon. So like you can make decisions assuming a bunch of those things are going to be true, which is like a lot of folks …”

“I don't think we've seen like across other product services, people actually invest in like real-time audio and real-time video and image stuff. And I think that's like the next iteration of the UX of how people are going to interact with AI models.”

“And I actually think if we've, if we fast forward like. 10 years, I do think there's going to be a lot of those experiences which like look eerily similar to the way that they do today, because it's just like so ingrained in like human culture, like how, like …”

“I think the future that we're going to see with the multimodal live API is really the ability for the model to or just like generally for models to be co-present in the experiences that you are. And like, you could imagine your IDE is the best experience of th…”

“GPT-IV was the first model that we trained where we could reliably predict the capabilities of that model beforehand. Based on the amount of compute that we were going to put into it, and you could actually, we did like a scientific study to show like, hey, th…”

“And I think with Gemini Three Pro, you can just, Vibe code games in real time, which is really, really awesome.”

“This is meant to showcase like, you know, single prompt, Generations maybe like a couple of multi-turn in some case, but like this isn't something that like we paid a team of like 10 engineers five months to come up with and then put it into our demo. Like thi…”

“So we have, like, 20 different simulators running at the same time. This was one-shot vibe coded as well.”

“I think for folks, again, who have used 2.5 pro, I think three point O pro is like roughly on par from a latency standpoint. So there's no I think in some cases it's actually, it's a little bit higher latency, but obviously the model is much smarter.”

“I think hopefully there'll be like some level of like similarity and you'll get some model agnostic infrastructure to help make that, you know, make developers feel a little bit easier about being able to move between models potentially.”

“Relative to today where you can literally just write a prompt and send images or videos to the model and have it do those tasks like with basically, you know, near or better accuracy than you would get from domain specific models is absolutely fascinating.”

“There's a delta in your output if you are AI assisted versus not across coding across every discipline right now.”

“We keep seeing this jump where the capabilities of the pro models get superseded by the flash models as the next generation comes. So if you look at our two point O flash model is actually on like every dimension, a stronger and better model than the previous …”

“And it feels like that is. More likely in the medium term going to be the thing that like continues to just like rapidly scale up relative to the other capabilities.”

“Jack Ray. Yeah. He's been a long time deep mind research scientist, was previously a pre-training person. We actually overlapped at open AI together a little bit, and then is now back at deep mind with no co-leading the reasoning effort.”

“And like, it feels like, again, like back to this, the thread around these capabilities, like it feels like long context with reasoning is like finally going to be that thing where like, it actually just like blows the lid off of it. And like, it makes the use…”

“If every browser in the world doesn't have this experience in the future, I'd be surprised if every IDE in the future doesn't have the ability to like, here, let me just show the model what's happening here, share my screen, talk to it live. Like, I think that…”

“The API keys actually by default are also free, so you can get, like, 1.5 billion tokens using Gemini across a bunch of different models for free today.”

“This will happen with text models. You can imagine a world where you go into ChatGPT and you say, write me a blog post about AI. It automatically will go and be like, let me generate a much higher fidelity description of what this person really wants, which is…”

“Like I think monetization when it comes to the store later this quarter, I think is going to be extremely exciting. Like when people can get paid based on who's using their GPTs, that's going to be a huge unlock and like open a lot of people's eyes to the oppo…”

“You're going to have an edge on other people. If you're providing AI that's not accessible in a chat bot, like people are using a ton of chat and like, it's a really valuable service area. Like it's clearly valuable because people are using it, but I think pro…”

“You're going to be able to build apps using Gemini three pro for free in AI studio.”

“It's available across a bunch of Google services like the Gemini app. If you're sort of looking for an assistant or the API for developers and enterprises, if you want to build this experience or build Gemini three into your own products.”

“That is all vibe coded just using the camera out of the box, which is a ton of fun.”

“We're talking about Nano Banana, AKA Gemini, 2.5 flash image, which is our new Gemini state of the art image generation and specifically image editing model which folks are loving and no pun intended going bananas for right now.”

“It's also only like, I think it's like roughly four cents for an image to be generated too. So it's like You know, you can let people go wild, and you're not going to break the bank, which is really great. It's like a thousand images is 40 bucks.”

“This is completely vibe coded. You can see this is literally one shot as well.”

“All of this experience that we've looked at so far is completely free. So you can do all this, like, you know, ideally you build a great product with this and you end up using the Gemini API and all that stuff, but like, you don't necessarily need to. The expe…”

“Explicit caching is nice. Like there's definitely use cases where it makes sense, but people want implicit caching. So I'm happy passing the cost saving on to developers. You don't have to do anything. It just works right now and you're saving money.”

“Cost of AI down 99% over the last two years.”

“So it went from seven and a half cents per million tokens to 10 cents. The sort of way that this was offset was we used to distinguish. I don't know if folks are familiar with this, but we used to distinguish based on input token volume. So it was like over a …”

“We have our reasoning model available for free for developers. You can use it in AI studio. You can go and get an API key and use it as well.”

“So this is a capability that is like Baked into the model itself, where it's like really able to deeply understand different objects and the ways in which they're sort of visually represented.”

“So we're showing these thoughts in the UI. If you were using the API, you actually wouldn't see these thoughts. It's sort of abstracted behind the scenes, but we showcase it in the UI just to sort of give you an intuition of what's happening.”

“OpenAI had been pushing for a really long time since ChatGPT came out and that was supposed to be like the first, one of the first weeks that like the whole company had like taken time away to like actually reset and have a break.”