Modal scored 5.1/10.

The page has no single differentiation claim it commits to. The hero says “AI infrastructure that developers love,” the feature section says “designed to help AI teams deploy faster,” the platform section says “build robust, scalable data applications,” and the footer CTA says “ship your first app in minutes.” Each of these is a different promise — developer affinity, speed, robustness, ease of onboarding — and none of them is the same promise. The buyer who lands here can’t answer the question “why Modal instead of Replicate or RunPod” because the page never answers it. Modal’s actual defensible claim is buried in the subheadline (“sub-second cold starts, 100x faster than Docker”) and never repeated or defended anywhere else on the page. Pick one claim — the cold start and speed story is the sharpest and most falsifiable — and make every section on the page a proof point for that one claim.

The testimonials are doing the wrong job. The page has more than twenty pieces of social proof, which is a significant asset, but almost all of them are individual engineers expressing enthusiasm (“brings me joy,” “magical,” “sooo nice”) rather than named teams describing a specific outcome at scale. The four named testimonials at the top — Brian Ichter, Mike Cohen, Aakash Sabharwal, Anton Osika — are the strongest signals on the page, but they’re formatted identically to the Twitter ticker below them, so the buyer’s eye treats them as the same category of evidence. The buyer evaluating Modal against Lambda Labs or Anyscale isn’t looking for joy; they’re looking for a team like theirs that ran a workload like theirs and got a specific result. Separate the four named testimonials into a dedicated proof section above the product grid, give each one a one-line outcome summary (“reduced inference latency from X to Y,” “scaled to 10,000 containers without ops overhead”), and cut the Twitter ticker to five of the most outcome-specific quotes.

The page presents five distinct products — Inference, Training, Sandboxes, Batch, Notebooks — without telling the buyer which one to start with or which one is the core. A buyer who arrived from a search for “serverless GPU inference” lands on a page that immediately asks them to also consider training, sandboxes, batch jobs, and notebooks. This is a navigation problem disguised as a product problem: the page is structured as a product catalog, not a conversion surface. The buyer who came for inference doesn’t need to know about Notebooks on the first visit; they need to know that Modal solves their inference problem better than the alternative they’re currently using. Restructure the product section to lead with the primary use case (inference, given the market and the ICP), treat the other products as “also available when you’re ready,” and add a single outcome-specific CTA after the inference description rather than a generic “Learn more” on each card.

The page never names what the buyer is switching from. Every testimonial that mentions a comparison — “compared to Docker, Cloud Run, Lambda,” “if you are still using AWS Lambda instead of Modal,” “similar to using Vercel for the first time” — is buried in the Twitter ticker where it’s easy to miss. The buyer evaluating Modal almost certainly came from one of three places: they’re running on Lambda and hitting cold start limits, they’re managing Kubernetes and drowning in ops overhead, or they’re on Replicate and hitting pricing or flexibility walls. The page never speaks to any of these situations directly. Add a short section — three columns, each naming a specific prior state and the specific Modal outcome — that makes the switching case explicit. “Running on Lambda? Modal cold starts are under a second. Managing Kubernetes? Define your infra in Python, no YAML.” This is the highest-trust move the page isn’t making.