Cerebras Names Claude Sonnet 4.5 on Its Homepage. Here's the Design Case.

The ABWatcher blog

Cerebras Names Claude Sonnet 4.5 on Its Homepage. Here's the Design Case.

Cerebras runs a direct, named-competitor bar chart on its homepage. We break down the layout choices that make it one of the rarest social-proof patterns in CRO.

Aiko Tanaka

Design Director · · Updated

TL;DR

  • Detected 70% rarity, 65% importance — a named head-to-head benchmark chart, logged on cerebras.net home.
  • Surface: layout, category: social_proof, page type: home.
  • Comparison: GLM-4.7 on Cerebras vs. Claude Sonnet 4.5 on GPUs, across three axes.
  • Metrics charted: Coding Ability, Output Speed (Tokens/s), Cost ($/M Input Token).
  • Pattern rarity (70%) signals most SaaS brands still avoid naming competitors directly in hero-adjacent modules.

What does the Cerebras homepage actually show?

A three-metric bar chart sets GLM-4.7 running on Cerebras hardware directly against Claude Sonnet 4.5 on GPUs, bar for bar. The evidence quote is blunt: "GLM-4.7 on Cerebras vs. Claude 4.5 Sonnet ... Coding Ability, Output Speed (Tokens/s), Cost ($/M Input Token)." There's no hedge language, no "up to," no asterisk-laden disclaimer bar underneath. It reads like a spec sheet, not a slogan.

That's the first thing worth sitting with before judging the layout: the content itself is a position, not just a design choice. Most B2B homepages dodge named comparisons entirely, routing that kind of claim to a gated competitor-comparison page three clicks deep, if it exists at all. Cerebras put it in the layer of the page that gets the most eye traffic on earth — the home surface, in a social-proof slot that normally carries logos or quote cards.

Why does the eye land on the bars first?

Bar charts exploit preattentive processing — length and color difference get read before a visitor consciously parses labels, which is exactly what makes a comparison chart faster to "get" than a paragraph of claims. In an F-pattern scan, the home page's top band usually front-loads headline and CTA; a chart embedded in that zone competes with the primary value prop for first-glance real estate. That's a deliberate hierarchy bet — Cerebras is willing to spend top-of-fold attention on proof rather than pure aspiration copy.

Three metrics, not five or eight, is also a hierarchy decision. Coding ability, speed, cost — these map almost exactly to the three questions a technical buyer asks in sequence: can it do the job, how fast, what does it cost. Gestalt's law of proximity does the rest; grouping the three bar-pairs tightly with shared axis labels tells the eye "these are one comparison," not three separate facts, which keeps the chart legible as a single unit even at a glance.

Is naming a competitor actually good CRO, or just bold marketing?

It's both — named, falsifiable comparisons remove the ambiguity that vague superiority claims leave behind, and ambiguity is friction. Most "trust" CRO advice centers on testimonials and logos; social proof placement guidance even recommends moving trust signals up near the form because proximity to the decision point matters. Cerebras skips the soft version of proof and goes straight to numeric evidence, which tracks with broader CRO research that numerical evidence is powerful for conversion because it gives anxious buyers something concrete to hold onto instead of a trust-me adjective.

The risk, obviously, is falsifiability cuts both ways — a named chart invites scrutiny a logo wall never does. That's precisely why it's rare. Most teams calculate the legal and reputational cost of a competitor rebuttal isn't worth the conversion lift. Cerebras made the opposite bet, and the fact that it's sitting on the home page rather than buried suggests internal confidence in the numbers holding up under a side-by-side audit.

How does this compare to standard social-proof placement?

Pattern Typical placement Specificity Rarity (per ABWatcher)
Customer logo wall Below hero Low (no numbers) Common
Testimonial quote card Mid-page or near form Medium (anecdotal) Common
Aggregate stat ("10,000+ users") Hero or nav bar Medium (numeric, vague context) Occasional
Named competitor benchmark chart Home, hero-adjacent High (falsifiable, three axes) Rare — 70%

The further right on this table, the higher the design confidence required — you need a chart component, not just a text block, and you need a legal sign-off most marketing teams won't pursue.

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

Treat the chart as a hierarchy exercise before a legal one: pick the two or three metrics your buyer actually weighs in sequence, group them with tight proximity so they read as one unit, and place real numbers where your eye tracking shows first-glance attention — not three scrolls down. If you're not ready to name a competitor, the same bar-chart logic still beats a vague superlative; a well-labeled self-comparison ("before/after," "us vs. category average") borrows the same preattentive speed without the legal exposure. Per value-proposition testing guidance, the strongest claims make a visitor think "this was built for someone like me" — a chart that answers their exact buying criteria does that work in under two seconds.

This sprint: audit your own homepage's social-proof module and ask whether it's giving visitors an adjective or a number. If it's an adjective, you have a test to run.

Common questions

What makes this detection rare?

ABWatcher logged it at 70% rarity because named head-to-head competitor charts almost never appear on a homepage's primary social-proof surface — most brands keep comparative claims off home entirely or soften them into unnamed "vs. traditional GPUs" framing instead.

Why does naming Claude Sonnet 4.5 specifically matter?

Naming a specific, well-known model makes the claim falsifiable and therefore more credible to a technical buyer who can verify it, unlike generic "faster than the competition" language that carries no accountability and gets discounted by skeptical readers.

Could this chart hurt Cerebras if numbers shift?

Yes — a named, dated benchmark is a liability if the competitor ships an update that closes the gap, which is likely why these charts are rare; most teams prefer vaguer claims specifically to avoid needing to update or retract a public number.

What's the CRO lesson for non-AI SaaS brands?

Specificity beats sentiment. Whether it's load time, cost per seat, or setup time, a three-metric, clearly labeled bar comparison against a named or category benchmark will usually out-convert an unnamed superlative, per standard trust-building CRO guidance.

Where on the page should a comparison chart like this sit?

Hero-adjacent, inside the first scroll, is what Cerebras chose — it trades some headline attention for immediate proof, a reasonable trade for a technical, skeptical buyer persona that wants evidence before narrative.

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.