Create a landing page for Baseten, the AI inference platform, that is demonstrably superior to the current homepage at https://www.baseten.co/ — not a clone, but a redesign that keeps the real product, real content, and real brand identity while raising the bar on visual craft, motion, hierarchy, and interaction quality. A visitor should be able to scroll the whole page and feel that this version is sharper, more confident, and more memorable than the live site, while still being unmistakably Baseten.
Download the real assets from the live site and use them locally: the Baseten logo and wordmark, the customer logos in the marquee (Abridge, Clay, Cursor, Decagon, Descript, EliseAI, Gamma, Harvey, HubSpot, Lovable, Notion, OpenEvidence, Parallel, Poolside, World Labs, and any others present), the model library logos (Kimi, DeepSeek, Z AI / GLM), the customer headshots used in testimonials, and any product or platform imagery that improves the result. Supplement with additional downloaded or generated assets where the redesign calls for them — backgrounds, textures, iconography — but every real customer logo and headshot should be the genuine file served from the site's CDN, stored locally so the page works offline.
Preserve the substance of the real page, reorganized and elevated:
- A hero built around the real positioning — "Inference is everything" — with the supporting line about the fastest model runtimes, cross-cloud high availability, and seamless developer workflows powered by the Baseten Inference Stack, and the two real calls to action ("Get started", "Talk to an engineer"). Give this hero a presence the current site lacks: consider a living visualization of inference itself — token streams, latency traces, request flows across regions — rendered crisply and performed smoothly, not a stock gradient blob.
- The customer logo marquee, with the real logos, presented with better pacing, contrast handling for dark-on-light marks, and a more considered treatment than the live page's static strip.
- The product platform section: Dedicated Inference for high-scale workloads, pre-optimized Model APIs (with the real model cards for Kimi K3, DeepSeek-V4-Flash, GLM-5.2 Fast and the library link), Training with the Loops SDK, and Baseten for Model Labs.
- The "fastest inference takes more than GPUs" section with its four pillars: bleeding-edge performance research (custom kernels, decoding techniques, advanced caching), inference-optimized infrastructure (any region, any cloud, blazing-fast cold starts, 99.99% uptime), DevEx built for rapid iteration, and Forward Deployed Engineers.
- The deployment section — "Scale fast — in our cloud or yours" — contrasting Baseten Cloud (fully managed, global, massive horizontal scale, single-tenant options) with Self-hosted (the managed-service experience inside your own VPCs, with optional hybrid flex capacity).
- The modalities section — "Engineered for the most demanding Gen AI apps" — covering rapid image generation, optimized transcription, SOTA text-to-speech, performant LLM runtimes, the fastest embeddings (BEI's 2x throughput and 10% lower latency), and ultra-low-latency compound AI with Baseten Chains (6x better GPU usage, latency cut in half).
- The testimonial section with the real quotes and real headshots: Nathan Sobo (Co-Founder, Zed Industries), Sahaj Garg (Co-Founder and CTO, Wispr), Jagath Jai Kumar (Full Stack Engineer, OpenEvidence — including the 160-millisecond latency detail), Mahendan Karunakaran (Head of Mobile Engineering, ClickUp — the sub-300ms transcription quote), and Waseem Alshikh (CTO and Co-Founder, Writer). Give these voices a stronger editorial presentation than the current card grid.
- A closing call to action — "Explore Baseten today" — and a complete, well-organized footer reflecting the real sitemap (Product, Platform, Deployment options, Modalities, Developer, Resources, Legal), with the status indicator, social links, SOC 2 Type II and HIPAA compliance marks, and the copyright line.
Throughout, the redesign should make deliberate, opinionated choices: a distinctive typographic system with real hierarchy instead of uniform weights; a color and surface language that feels like infrastructure-grade precision rather than generic SaaS; motion that responds to scroll and pointer with purpose — the marquee, the hero visualization, section reveals, hover states on cards and links — and none of it gratuitous; visible latency, throughput, and scale numbers treated as first-class design material, because for this audience the metrics are the message. Every interactive element needs hover, focus, and active states; the page must be fully responsive from small phones to wide desktop; and it should remain fast and legible with motion reduced.
Do not take shortcuts, do not produce a cookie-cutter SaaS template with Baseten copy pasted in, and do not stop at a recognizable approximation of a landing page. This skill imposes no token budget limit, so pursue the full depth of the experience: real downloaded assets, complete sections, thoughtful responsive behavior, considered motion and micro-interactions, accessible markup, and the small details — favicons, meta tags, link affordances, empty and edge states where relevant — that separate an authored page from a generated one. Keep refining until placing this page beside the live baseten.co makes the live site feel like the older draft.