year basis more or less the same but I don't see acceleration Q4 compared to Q3 okay perfect thank you thank you very much our next call comes from the line of Robert Kulberth of Evercore Robert your line is open hi two questions please as partners use base before in place of Wix studio just wonder if you talk about attach rate of Wix business solutions on a headless spaces. Maybe also just some early learnings from some of the headless initiatives you have with the LLM partners as well. And then just wanted to double-click on the GM profile in H2. I think you said one to two points of benefit, H2 over H1 from base 44 or the base 1 model specifically. But then you also talked about some AI savings in the rest of the business. So just wondering if you were to put that all together, maybe you could give us a consolidated view of the GM improvement in H2.

I will start with the gross margin and about the two points that you just mentioned. And then Avishai will continue with the first one, O'Neill. So with regard to the gross margin improvement in the second half, I said that it's going to improve by two points. And this is mostly coming from the savings that we see compared to the guidance we provided previously about Base 44, meaning that the improvement in gross margin and the usage of the model of Base 1 and the fact that it's, as Avishai mentioned, that it's even performed better. So, we see that the usage of base 1 is even much higher and better than what we expected and faster.

Therefore, we updated the gross margin as a result of that, meaning that we see a dramatic change in gross margin of base 44, and it's translated to a two-point improvement in the consolidated revenue in the consolidated gross margin a roll for for the first part of your question so you know as I wish I mentioned before you know partners you're asking about partners using base 44 instead of studio you know the dynamics here is I wish I said about you know partners and base 44 are very very early and it's very you know I think it's it's too early to comment about whether it's supplementing or it's actually additional. By the way, we are seeing some that are using both for different use cases. And I think that as time progresses, we'll probably have segmentation more and more of some partners and agencies that are sticking to one product, other that are using different products for different kinds of projects that they're doing. So I think it's early in this cycle. It's definitely an evolution of this segment, and it should be very interesting. Our goal is to try to be there and deliver the right value on, you know, whether they're trying to go one way or the other. By the way, to some extent, and the second part of your question was about headless, it also applies to headless. Obviously, in a world where there are more and more agentic solutions out there, we want people to be able to benefit from the value of the business stack that's on Wix in a very easy manner. And this is what Headless is all about. It's something we're starting to see pick up, but, again, it's an early cycle. Obviously, it's aimed towards people who are more professional in their youth and what they're trying to build. And we're seeing traffic coming to this from various places over the Internet where the professionals are dealing and working with a genetic solution. You know, it can be, you know, things that are coming from a topic, open AI, our own stack. And our goal is to definitely be there and add that to the portfolio of solutions that we offer the professional crowds. I do think that in the coming few quarters, it's going to become more and more clear exactly what is the preferred path for each and every one of them.

Great. Thank you.

Thank you very much. Our next question comes from the line of Naveed Khan of B-Riley Securities. Naveed, your line is open.

Great. Thank you very much. So I understand you guys are not updating the ARR for Base44, but just curious if the trajectory is similar to what we saw between your last few updates. And are you still taking share from the level in the U.S.? you still ahead there because i think that's something you pointed to before so that's my first question and then uh in terms of just say base 44 mix of monthlies versus annuals can you just maybe talk about how that next looks like uh seems like you're getting more annuals versus monthly but any color they would be helpful thank you hey uh so for the the all uh as we said before this is not a KPI where we intend to report it's you know it's trending in similar manners as before in terms of what you asked about the mix of monthly

and annual then yes I would say there's more adoption of annual but we are not in the in the place where we want to break it out to the exact numbers right now for the lovable market share in the US you want to comment on that Yeah, I think that it's important to say we are not taking shares from Lavable.

We are inventing a new market together with Lavable. And we are competing on the percentage of that going forward. And I think it's mostly us supporting each other more than competing with each other by educating the market and educating people that the fact that they can now build applications themselves and you know invent products and actually make that and manifest that in reality which was she is shocking for most people so we I would say we will most support each other than competing with each other yeah thank thank you guys yeah I must talk I think you had said that you were ahead of level in terms of taking the share but not necessarily about what I didn't share hard if If you're trying to find the precise market split between us to Lavable, it's extremely hard. They are a private company. They report numbers based on metrics that are not clear. Okay, so it's very hard for us to distinguish marketing on their numbers from real numbers. and I can say that our estimation is that in the more important market the United States we are ahead of them but and to say by how much is again tricky we can only look at secondary metric says that we can measure on the internet I would say we're in a very good place okay thank you guys thank you very much Our next question comes from the line of Andrew Boone of Citizens.

Andrew, your line is open.

Thanks so much for taking the questions. I wanted to go back to base one. Can you just talk about the improvements of the model? If we think about the trajectory of base one, should we really think about a parallel path with open source more broadly? So as we think about Kimmy K3, if we think about other open source models that have come to market, is that the right trajectory of what we see for improvements for base one or how do we think about that? And then just connect that more broadly to what we're seeing in terms of conversion or retention. How do we think about the improvements of models with base 44 improving the overall business? And then secondly, just more of a bigger picture question. A lot of people think about frontier models and their progress and especially just the competitive threat from cloud code or other frontier type models how do we think about just the competitive mode broadly speaking for a base one and open source more broadly versus what is the frontier as we think specifically about cutting assistance thanks so much all right so those are pretty different questions the fourth one is base one will it improve in parallel with open source model so I want to be saying first of all the answer is yes of course however I don't think that that is the most significant part of the improvement all right for

most of the applications that you want to build today Kimi free GLM 5.2 5.4 thing and and clothes or a 5 opus 5 or a or the latest codecs they are all good enough for most applications right because most of us don't do those complicated things that require that extra a few percent in model intelligence right so for most of the application the improvement that we need to make on base one how to build those applications better okay and it's a different thing it's not about how we make it more intelligence in in Hungarian poetry right which is one of the things that or on any other of those finger molecular biology right all those things that those frontier models are trying to push the envelope a lot of what we need to do is how we make applications that people prompt and a lot of the time those people are not professional developers and they don't prompt them in a professional way right they're lacking a lot of of the definitions they don't put all the information that you need how do you make those base I'll make base one better than solving those specific problems and so yes we will benefit from continuation of development in the open source environment absolutely but the vast majority of the value will come for what we do and not from implementing the open source models how does base one improved impact conversion and retention I believe that was your second question well obviously if you come to Wix and to base 44 and you prompt something and you get a bad application you're going to convert a lot and you're going to get a lot less than if you went to base 44 type the same prompt and get a better application. Even if you don't get a perfect application, but in a place that you can now feel confident that if you start working and continuously to prompt, you'll get to what you want, that's already a dramatic improvement in your chances to convert. And we see that, we measure that, we have now a year of consistent measurements on that. We know that that's the case. So, by improving base one, I'm confident we will continuously improve base 44 conversion, which is why I'm so excited about it. The other side of it, of course, is exactly the same for retention. What is the competitive mode for base one and open source broadly based on team models? That's a very long question. I think that it's beyond the scope of this conversation. I'll just say my two cents and currently everybody is using exactly the same algorithm okay we yes we do some engineering modification on top of it and and but but that's pretty much it so the mock is not huge okay for frontier models however they've proven in the past that they have the ability to innovate even within the same algorithm which is attention and based on farmers right as an algorithm, and create new things. So if they continue to do that, they'll have some kind of a model. There is another side to that, is that there is a level where it doesn't matter or legal action or government action will start preventing the evolution and release of new models. We see that we meet this, right, on security, cybersecurity issues. So I think that between, and the fact that for a lot of the times, the current model already intelligent enough so if you're a commercial model a frontier commercial model but for the task that your customers need that extra intelligence that you've added are not significant okay it's very hard to justify prices in other words it's a very interesting conversation I think beyond the conversation of our discussion body but I think it's It's extremely important for the Western world to have really strong frontier models in the Western world, and I really hope that those guys at OpenAI and Anthropic and Google will continue to innovate and drive fantastic products. Thank you.

Thank you very much. This concludes our question and answer session. Thank you for your participation in today's conference. You may now disconnect.

Learn better.