Baseten Ditched Feature Bullets for Named-Model 'Try It' CTAs

The ABWatcher blog

Baseten Ditched Feature Bullets for Named-Model 'Try It' CTAs

Baseten's hero swaps generic copy for live "Try it" buttons on named models like Kimi K3 and DeepSeek-V4-Flash. Here's why that's a smarter CRO play for infra products.

Diego Hernandez

Growth Marketer · · Updated

TL;DR

  • Brand: Baseten (baseten.co), home page, hero surface.
  • Detected pattern: Named model "Try It" buttons replacing generic feature bullets.
  • Models called out by name: Kimi K3, DeepSeek-V4-Flash-0731, GLM-5.2 Fast.
  • Rarity score: 70% — this pattern shows up in a small slice of the brands ABWatcher tracks.
  • Importance score: 60% — meaningful hero-level change, not a cosmetic tweak.
  • Category: interactive_demo — the CTA leads to a live test, not a form.

What did Baseten actually change in the hero?

Baseten replaced the standard "here's what our platform does" bullet list with three named models, each getting its own live "TRY IT >" button. The evidence pulled straight from the page: "Kimi K3 / TRY IT >, DeepSeek-V4-Flash-0731 / TRY IT >, GLM-5.2 Fast / TRY IT >."

No "Fast inference at scale." No "Deploy any model in minutes." Just the actual models a visitor might already be searching for, each with a direct path to run inference before creating an account.

That's a meaningful shift in what the hero is optimizing for. Most infra and dev-tool heroes sell the platform. This one sells the product's output, instantly, by name.

Why does naming specific models beat generic feature copy?

Because technical buyers don't trust adjectives — they trust proof, and naming the exact model they came to evaluate removes a layer of translation.

If you're a developer landing on Baseten because you heard DeepSeek-V4-Flash is fast and cheap, a bullet that says "optimized inference infrastructure" makes you do work. You have to translate that claim into "okay but does it run the model I actually want." Naming the model does that translation for the visitor. It also signals something subtler: Baseten is current. Shipping GLM-5.2 Fast and a dated DeepSeek checkpoint (V4-Flash-0731) in the hero says "we track the frontier," which matters a lot in a market where model releases move faster than most companies' marketing sites.

This lines up with a core CRO principle that keeps showing up across the research: personalization and specificity consistently outperform generic messaging, because a one-size-fits-all pitch never resonates as hard as a message that mirrors exactly what the visitor already has in mind.

Takeaway: if your buyer searches for things by proper noun — model name, integration, library, competitor — put that noun in the hero, not a category description of it.

Does an interactive demo actually convert better than a signup form?

Yes, when the product's core value is something you can experience in seconds — letting people touch the real thing beats asking them to trust a claim.

A "Try It" button that fires live inference is a demo with zero setup cost. Compare that to the usual SaaS hero pattern: "Start Free Trial" → email gate → onboarding flow → finally, you see the product. Baseten collapses that funnel. The product sells itself in the first five seconds, before any account exists.

This is the interactive_demo category doing exactly what it's supposed to do. It's rare — 70% rarity in our detection — because it's genuinely harder to build. You need live inference endpoints sitting behind marketing buttons, not just a video or a static screenshot. Most teams default to video because it's cheap. Baseten chose infrastructure-as-marketing instead.

How common is this pattern across the sites ABWatcher tracks?

Rare. Named-entity CTAs in a hero section are still an edge case compared to generic demo requests or newsletter captures.

Pattern Typical hero CTA Friction before value
Standard SaaS hero "Start Free Trial" / "Book a Demo" Email + onboarding
Feature-bullet hero "See Features" Scroll + read
Baseten's pattern "Kimi K3 → TRY IT" Zero — instant inference

The middle row is still the majority pattern across ecommerce and SaaS homepages we monitor. The bottom row shows up almost exclusively in developer-facing infra products where the "feature" and the "product" are the same thing — you can't fake trying a model.

What should growth teams steal from this for their own roadmap?

Don't bury your most specific, most-searched-for asset under generic category language — surface the named thing people already want, and let them act on it immediately.

If you sell integrations, name the integrations in the hero, each with its own CTA. If you sell templates, name the templates. If you sell models, name the models. The Unbounce CRO playbook makes the same point from a different angle: specificity and fast iteration beat polished generality, and you should be testing headline and CTA variants against real behavior instead of guessing what resonates.

The bigger lesson from Baseten's hero isn't "add more buttons." It's that the fastest path to conversion is often removing the translation layer between what the visitor wants and what they can click.

Common questions

What is a "named model" CTA pattern?

It's a hero design where instead of generic feature bullets, the page lists specific named products — in this case AI models like Kimi K3 or GLM-5.2 Fast — each paired with its own direct action button. It skips abstract platform messaging in favor of concrete, searchable entities visitors already recognize.

Why did Baseten choose live inference buttons instead of a signup form?

Baseten's core value proposition is model inference speed and availability. A live "Try It" button lets visitors experience that value directly, with zero signup friction, which is a stronger proof point than copy describing performance claims a visitor can't verify themselves.

Is this pattern only relevant to AI infra companies?

No. Any product with a well-known catalog of named items — integrations, templates, plugins, datasets — can apply the same logic. Name the specific thing your visitor searched for and attach a direct, low-friction action to it instead of hiding it behind a generic "Explore Features" link.

How rare is this CRO pattern across tracked sites?

ABWatcher logged this detection at 70% rarity, meaning it's uncommon compared to standard demo-request or trial-signup CTAs. Most companies default to generic CTAs because building live, interactive demos behind marketing buttons requires more engineering investment than a static page.

What's the one thing to test this sprint?

Audit your hero copy for generic category language — "powerful," "enterprise-grade," "all-in-one" — and replace at least one bullet with a named, specific entity your buyer already searches for, paired with a CTA that removes a step rather than adding one.

Read next

More from the ABWatcher blog

From the glossary

Vocabulary the article uses. See all terms →

  • Shrinkage Estimator— A statistical adjustment that pulls noisy per-variant or per-segment estimates toward an overall average to reduce overfitting to randomness.
  • Flicker Test— A diagnostic check to confirm whether users briefly see the original page before a variant renders in client-side experiments.
  • Pre-Experiment Bucketing Audit— A review process confirming users are being randomly and evenly assigned to experiment variants before launch.
  • Above-Fold Exclusion Rule— A test-design rule that deliberately prevents changes to above-the-fold elements so an experiment isolates below-the-fold effects.
  • Flicker-Free Testing— A technical setup that prevents users from briefly seeing the original page before a test variant loads.
  • Orthogonal Test Design— A test structure where multiple variables are changed independently so their individual effects can be measured without interference.

See more like this

ABWatcher catches A/B tests like this across 1,000+ brands.

Watch live experiments at 1,000+ high-converting brands, complete with hypothesis and takeaway. 7 days of Pro, free, no card, then Free forever for ten watched companies.