AI Mornings with Andreas Vig

Anthropic's $965B Valuation & Protein AI Beats AlphaFold

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 5:41
Today's top AI stories: Anthropic raises $65B at $965B valuation, Claude Opus 4.8 launches with dynamic workflows, CZI Biohub releases open-source protein model surpassing AlphaFold3, and more including the mystery of Hy3's popularity, Apple's leaked Siri plans, and Visa's agentic payments investment.
SPEAKER_00

Hey, welcome to AI Mornings with Andreas Vig. It's Thursday, May 29th, 2026. Anthropic just closed what's likely the biggest private funding round in tech history. The company raised 65 billion US dollars in a Series H round, reaching a $965 billion valuation. That makes Anthropic the most valuable AI startup in the world, surpassing OpenAI. Claude now generates over $47 billion in annual revenue. The round was led by Altimeter, Dragoneer, Greenoaks, and Sequoia, with Amazon and several other major investors participating. This is probably the last private fundraise before an IPO that'll be closely watched. And there's more from Anthropic. Just 41 days after releasing Opus 4.7, they've launched Opus 4.8. The new model shows improvements across coding benchmarks and agentic tasks, but the real news is what comes with it. Dynamic workflows lets Claude Code spin up hundreds of parallel subagents to tackle large-scale migrations across entire code bases. There's also a new effort control that lets users choose how hard Claude thinks about a response, from low to max. Fast mode is now three times cheaper than before. Early testers say Opus 4.8 is four times less likely to let flawed code slip through and noticeably more honest about its uncertainties. Same pricing? $5 per million input tokens, 25 per million output. In research news, the Chan Zuckerberg Initiative's BioHub just released something they're calling a world model of protein biology. It's a fully open source stack for predicting and designing proteins. The centerpiece, ESM Fold 2, claims to beat Google DeepMind's Alpha Fold 3 on structure prediction and protein-protein interactions. They've already used it to design binders against five cancer and immune targets, with lab hit rates between 36 and 88%. The ESM Atlas maps 6.8 billion protein sequences and over a billion predicted structures. Everything is MIT licensed and free. This could significantly accelerate drug discovery. Something strange is happening over at OpenRouter. A model called High 3 Preview from Tencent has quietly become one of the most used models, even surpassing Claude in token usage. The weird part, benchmark tests show it's not particularly good and hardly anyone's talking about it. It's served by only one provider, a Singapore company called Silicon Flow, and it has a restrictive license. One analyst dug into the data and couldn't find a clear explanation for its popularity. DeepSeek 5.4 Flash actually offers better effective pricing. If anyone knows what's driving high-three usage, I'm genuinely curious. The OpenAI Foundation, the nonprofit arm that owns 26% of OpenAI's for-profit business, committed $250 million to address AI-driven economic disruption. The funds will go toward understanding AI's economic impact, retraining workers, and building long-term security through ideas like tax shifts from labor to capital and sovereign wealth funds. First initiatives will be announced later this year, though some argue the urgency doesn't match the speed of current layoffs across industries. A new startup called Trajectory launched with $15 million in seed funding to build AI that keeps learning on the job. Founded by researchers from DeepMind and Apple, the platform captures user corrections and edits to continuously post-train models. Right now they update weekly, but they're targeting hourly updates or even learning at every interaction. Early customers include Clay, Harvey, Decagon, and Rogo. The idea of models that compound in quality from real feedback is compelling. Sesame, the conversational AI startup from Oculus Founders, released its iOS app with four AI agents named Maya, Miles, Simone, and Charlie. Each has their own personality and memory. What's interesting is how they handle conversation. The AI can pivot mid-sentence as it retrieves new information, kind of like a human remembering something important. A million people tried the research preview earlier this year. An Android version is coming, with smart glasses planned for 2027. Visa invested in Replit to work on agentic payments. They're exploring how AI agents built on Replit can accept payments directly using Visa's trusted agent protocol, a system for agents to securely identify themselves. Over a thousand Visa employees already use Replit internally. Replit also launched self-serve enterprise access for contracts up to $200,000 without needing to talk to sales. Finally, Bloomberg published leaked renders of Apple's AI plans for iOS 27. Siri will now emerge from the dynamic island, and swipe down search will be powered by Google's Gemini under the hood. There's also a standalone Siri app to compete with ChatGPT and Claude directly. Apple is betting its 2.5 billion device install base can introduce AI to people who haven't adopted standalone tools. Full reveal at WWDC in June. That's all for today. Catch you tomorrow.