Base 44’s – Base44’s founder says building its own AI model is meant to avoid the “AI-slop” look and deliver faster, cheaper results. A side-by-side test against Anthropic’s Opus 4.8 shows Base 1 generating a vibe-coded e-commerce site sooner, while Opus 4.8 uses credits
A fictional mood lamp should be fun to build. But the real test came with the button press—choosing which AI would generate the website.
Base44’s CEO and founder. Maor Shlomo. has argued that the startup trained its own LLM to steer vibe-coding away from the “AI-slop look” that can define these tools. On paper, that means more than aesthetics. In practice. it turns into how quickly a site appears. how many credits it burns through. and whether the design feels like it came from a specific product—or just another frontier-model template.
Base44 is a San Francisco-headquartered vibe-coding startup. It competes with companies including Lovable, Replit, and Cursor. The company’s Base 1 model was tested against Anthropic’s Opus 4.8 by using the same prompt: build an e-commerce website for “Lumos. ” an emotional support lamp that reads Google calendars and changes its hue depending on the ongoing event. The lamp’s colors were specified as yellow for focus mode. blue for downtime. pink for meetings. and red for impending deadlines.
Design gurus have said websites built with frontier models—including Anthropic’s Opus 4.8. OpenAI’s GPT-5.5. or Gemini 3—often end up sharing similar interfaces and design elements such as rounded corners. beige color palettes. and excessive use of emojis. Paul Bakaus. the CEO of AI design startup Impeccable. described these sites in a June interview with Andreessen Horowitz as an “algorithmic Uniqlo or Ikea” — functional but basic.
Shlomo’s promise is that Base 1. in more advanced future versions. will “create something that looked uniquely different” every time—though he also said it is “going to take some effort. so we are not yet there.” He also said Base44’s model would produce faster results and cost users fewer credits than using frontier models. Base44 is a subsidiary of the website-building company Wix.
To run the comparison, the test used Base 1 as the working model while building the Lumos site. There was no instruction about website design—only the product concept and the requested color mapping—so the model would choose its own look. Before pressing Enter, the model choice mattered: free users do not have the toggle. Only users with the $40-per-month “Builder” subscription or higher can switch between AI models.
After a few minutes, Base44 generated the site. The first surprise was the color palette. The website used a deep blue background—already a departure from the lighter or beige-colored front ends commonly seen in vibe-coded sites—along with a bright yellow accent color. The rest of the layout was basic and still carried familiar vibe-coding cues: rounded corners and plenty of emojis.
Then the prompt went through the same process again, but with a different model. Opus 4.8 took longer to generate the website than Base 1, with Opus feeling like “twice as long” in the test. When it finally loaded, the first “aha!” moment was the return of the beige background.
Aside from the color palette, the two versions looked similar enough that the difference landed more as a matter of familiarity than function. Both used the same number of credits: 1.2 message credits each out of the 250 monthly credits.
The gap showed up again when both models were pushed to do the same incremental feature change. The test asked each site to add one extra set of color modes—green for exercise and teal for eye breaks. Opus 4.8 burned through 1.4 message credits to complete the task, while Base 1 finished in 1.2 credits and did it faster. The difference was small. but it was consistent with the argument Shlomo has made publicly: Base44 expects to be cheaper in credit terms and faster in turnaround.
Even so, the results did not deliver a sweeping redesign that would prove Base 1’s eventual promise today. Taken together, the test did not reveal a major functional difference in how the two models worked. The main visual difference was speed on Base 1’s side and. in terms of design feel. Opus looking a tad more generic.
Shlomo said the Base44 team will conduct “reinforcement learning” on Base 1. a process involving prompting the model to keep generating designs that look new and unique. In the meantime. this test left the clearest takeaway in the only place that really matters for builders—between waiting and clicking—and in how quickly design credit costs add up.
“We’ll check back in when the next version is released,” the test concluded, with the implication hanging in the air: Base44 is trying to outrun the beige-and-emoji sameness, but the current iteration still lives close to the same design gravity.
Base44 Base 1 Anthropic Opus 4.8 vibe coding AI design Wix subsidiary Builder subscription message credits reinforcement learning website generation Impeccable Paul Bakaus algorithmic Uniqlo or Ikea