The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
00:44:53 · Harry Stebbings, Rory O'Driscoll
[00:44:53] S2: Sorry. I didn't quite understand that analogy. It's it's a bit like saying, I wanna be in scale, not legora, because I don't know who's gonna win the application layer, but I do know that models are gonna need to get better. That's a better comparable, which would be understandable. [00:45:10] S1: Personally What what you're what you're saying implicit in that is you believe there's a statement in that that each robotics company has its own foundation model and the companies like we're in the journalist pie and skilled, like if journalist pie and skilled become the dominant foundation models for robotics, then it is equivalent to my foundation model comment. But if every robotics company has to build its own specific model, then yes, then it becomes a very fragmented market at the model layer. And I don't know, it's definitely one of the areas where I have a fair number of robotics investment and the least amount of clarity. [00:45:43] S2: Okay, boys. There's a lot else. Guide me in terms of where you would like to go. We can talk about Anthropic cutting the price of Haiku by 90% as the price war with OpenAI further intensifies. Jason, I'm I'm relatively intrigued to hear your thoughts on that given how intensely you use them.
00:46:01 · Jason Lemkin
[00:46:01] S0: I mean, look. I know we've talked a lot about this kind of stuff on the show. I I I will say it's under discussed. These models, Haiku five five and Sonnet five five, they're they're great. And they're really, from my perspective, the first thing that's actually been deflationary with LMs and AI. Nothing else has been deflationary because no matter how many you talk about token prices falling, we've been consuming so many more tokens. We're not getting the benefit of it, right? None of our portfolio companies are seeing their cogs fall in AI, are they? Not really, right? We've all consumed so many more tokens and used more and more of the state of the art frontier models that we haven't seen inflationary benefits. These to me are the first ones that are deflationary. Haiku basically, you know, since the beginning of it, since before the show, there's always been the N minus one model, which isn't quite the frontier one. And you get the last year's Cadillac and Fins for this year's shining. It was sort of worth it, But now, Haiku 55, which is a joke. I mean, developers would laugh at you if you used Haiku 55. It's almost as good for a lot of use cases as Sonnet. And it's a tenth or the twentieth the price. It's like shock. It's so cheap. It's almost... Haiku 55 is like the first reasoning model and you can dial up how much reasoning you want that you can almost not care about what it costs like JEV. We haven't had this before that worked for me. We haven't had a reasoning level model where you didn't have to stress every token, every dollar, because it is a tenth the price of what Sonnet 4.5 was. It's a tenth the price. So I think it's disruptive. Sonnet 5.5 is interesting as an aside. It's basically, you know, it seems at least in the testing that I've done, it's almost as good as the Opus models before. But what's interesting, it's only about 10% cheaper than the previous on it, but it's twice as fast. And these evals don't talk enough about speed. Like this is why a lot of these things are useless like Grok and others, cause they're too slow or they time out. Speed, like if you can have your chat go twice as fast, it's a big deal, right? And Sonify five is twice as fast, 10% cheaper, not that much and better. Speed is also deflationary because you can do more and more quickly.
00:48:02 · Jason Lemkin, Rory O'Driscoll
So for me, these two plus JEV, the first time we've seen deflation. And yeah, you're going to have your Jevan's paradox, so we're going to use even more of course, but this is the first time you could literally go in swap out models and have much more speed and pay a lot less, like in a week. It's exciting. And I think it's most exciting for like B2B stuff where we're all still making these trade offs for a lot of B2B apps, right? We're making it how much, how good can we make the resolution? How good can we resolve this issue? How good can we do this workflow? If it's twice as fast and in some cases attempt the price, we sure better see a lot of results at the next board meeting. No more excuses, boys. It's the first one at a practical level I've seen this deflationary. [00:48:42] S1: And I think the next interesting thing will be, you know, college... You haven't called it price elasticity because price elasticity simply says, you know, for a proportion reduction in prices, does the net dollar spend go up or down? The bet obviously they're making is it's highly elastic and it goes down and you end up spending more and we'll see. That's obviously the bet. I mean, you would not want to be cutting prices in a demand inelastic product, because then you'd have the classic problem of high fixed costs, low demand expansion and lower prices, which is terribly horribly bad. So the bet here, as you see these price cuts, is in fact that this stuff is very elastic in terms of demand.
00:49:21 · Jason Lemkin, Rory O'Driscoll
[00:49:21] S0: Yeah. And I assume we'll see it in the in the s one soon enough. Assume... Listen. All the accounting for these LMs is gonna be interesting. Right? But I assume you can, when you're using Haiku or Sana, when you're using essentially the N minus one or N model, the only cogs is going to be inference, right? You can basically write off the training costs because you can assign that all to the frontier models, to Fable and Opus, right? Assuming that's the case, it's sort of a win win all around, right? Because I believe that Anthropics inference margins are quite high, right? Before I believe it, but I believe their real world gross margins are still pretty crappy because of training costs, right? So I think it'll be interesting to see how they break this out. Right. But I think it's, I think it's exciting, but yeah, we'll just use more. There's going to be a lot of stuff where it's, it's like, you know, taking a twenty minute shower when you don't have to pay for the hot water, you never get out. Right. [00:50:07] S1: Just as a reminder, and I agree with Will, and I agree with that assessment, but when you cut prices by 90%, we're going to use more, it doesn't cut it. We got to use 10 times more. Otherwise, demand elasticity is less than one and you actually revenue goes down. [00:50:18] S0: I do think we'll use 10 times as much because what's going to happen instead of sampling something, you'll just do it all. [00:50:23] S1: That was what my colleague said on using JEV. You'll use it in every row of the database.
00:50:23 · Rory O'Driscoll, Jason Lemkin, Harry Stebbings
[00:50:23] S1: That was what my colleague said on using JEV. You'll use it in every row of the database. [00:50:27] S0: Everything. And the key is speed, Matt. We don't discuss speed as much. JEV is so fast, right? And, Haiku 5.5 is actually slower than the one before, but it's not even comparable. It sounds twice as fast. You could just use more of it because it's faster. If it's slower, it doesn't it doesn't really... You can only use so much in parallel. But, yeah, I agree with your part. You just use it for everything. [00:50:44] S2: What does this mean for open source in US enterprises? Does this make it inherently less attractive where you're like, hell, I can get Haiku and Anthropic for 90% less? That really reduces the attractiveness of open source.
00:50:44 · Harry Stebbings, Jason Lemkin, Rory O'Driscoll
[00:50:44] S2: What does this mean for open source in US enterprises? Does this make it inherently less attractive where you're like, hell, I can get Haiku and Anthropic for 90% less? That really reduces the attractiveness of open source. [00:50:56] S0: That's why I thought... I said on the... Well, I don't know whether it Dev or or Jack, but I said you could imagine it would all flip around next year. And then the whole open router thing would change and people are not loyal. No one wants to use right or wrong. No one wants to use at least China based open weights models. No one in the enterprise wants to use them. So when you have a domestic solution, ideally from a vendor that's already been qualified and trusted, I'm not saying it will happen because there's a lot of smart people working at DeepSeek at all, right? I mean, some of the best of the best, but you could imagine that this being a moment in time and in twelve months from now, they're niche models that are used by some companies we invest in, but really in the enterprise are almost nonexistent. Again, I think Salesforce making Finn rip it out before the deal closed is telling. It's a moment in time, maybe that won't happen. I just, don't people don't want these China open weight models. They just don't want them. [00:51:45] S1: But but they'll have reflection.