DigitalOcean

You can't trust a model decision you didn't measure on your own workload. Fable 5, GPT-5.5, an open-weight model, and your own router policy: they all look great on a leaderboard. None of that tells you how they do on your data, on your task. You find out how the models actually perform in production, when your users do. DigitalOcean Model Evaluations tests any candidate on your workload: your dataset, your judge, and your metrics. In this demo, see how Fable 5, Opus 4.8, and a custom Inference Router scored on a legal dataset. The result isn't what the leaderboard would predict. Don't guess. Evaluate.

Angela Hoover is building Andi, an AI search engine that actually answers questions, scaling a 15-billion-page index while filtering out spam so you get the real results. 🤯 Since joining the DigitalOcean Startups program, Andi has grown to 150,000+ monthly active users & recently launched their API. As a lean AI startup, they need infrastructure that just works & they’re eyeing DigitalOcean’s Inference Engine to power real-time answers at scale. In Angela’s words, "We’ve saved on cost & on speed to production. We want to grow with DigitalOcean & scale together." Angela, the feeling is mutual! 🤝

Today we’re excited to welcome three outstanding leaders to DigitalOcean’s executive team: ➡️ Kevin Van Gundy, Chief Revenue Officer – scaled Vercel 50X, deep roots in AI Infrastructure ➡️ Leo Leung, Chief Marketing Officer – launched inference platforms at Google Cloud, built OCI from 0 to $12B ARR ➡️ Brady Mickelsen, Chief Legal & Administrative Officer – seasoned public company operator from Tanium, C3 AI, and Oracle They arrive at a pivotal moment for DigitalOcean: Q1 2026 revenue up 22% YoY, AI customer ARR up 221% YoY to $170M, and 81% of AI customer ARR came from inference and core cloud, not bare-metal GPU rental. As our CEO Paddy Srinivasan put it: "This era needs a new type of cloud. We built it—and now we have the team to win it." The inference economy is real. The demand is real. And DigitalOcean is built for it.

Anthropic Claude Fable 5 is now available on DigitalOcean. 🚀🆕 Anthropic's most capable generally available model is built for autonomous knowledge work & complex, long-running coding tasks. Run it directly within your existing DigitalOcean environment alongside your apps & data for predictable usage-based pricing & fully managed infrastructure.

DigitalOcean is heading to London Tech Week. 🇬🇧 📍 Find us at Booth 183 for AI conversations, live demos & 1:1 meetings. 🍻 June 9: Kick off the week with Techstars. Cocktails, live music, & 150 founders, builders, & investors. 🧠 June 10: Join Jeff Fan's hands-on session: Deploy AI in 5 Minutes. 🕵️ June 11: Catch Jeff at The AI Summit, where he'll break down why production agents need more than GPUs.

Treating every inference request the same is lighting money on fire. Not all requests are equal. "Reset my password" doesn't need a frontier model. "Build a new feature" does. Routing them the same way means paying frontier rates for tasks a cheaper model handles faster and at a fraction of the cost. DigitalOcean Inference Router fixes this automatically—sending each request to the right model based on task complexity, with no changes to your code. The results speak for themselves. For example, in this debugging demo, using Inference Router delivered: • 96% cost reduction on routed requests • 50% faster time to first byte • 75% faster end-to-end Open-source engine. Drop-in API. Right model, every request. No quality tradeoff. Live since April, DigitalOcean Inference Router has been running in production, getting better with every open-source contribution.

Meet the team behind #DOdeploy. 👋 From producing a live keynote to demoing the Inference Router on stage to advising customers on GPU infrastructure in real time, Deploy was a cross-company effort, & a window into how DigitalOcean works every day. Three employees share what it looks like to build the AI-Native Cloud from the inside: ownership, speed, & shipping things developers actually use.

DigitalOcean Model Evaluations is now in public preview. 🆕🤖📊 Test frontier models, router configurations, & fine-tuned models against the same evaluation framework to compare quality, latency, & cost. Catch failures before your users do & ship with confidence, backed by real evaluation data.