Top 5 Stories

  • Alibaba unveiled Qwen3.8-Max, a 2.4-trillion-parameter model with a million-token context window that the company says can code autonomously for weeks. This escalates China’s push to close the AI gap with US rivals. — CNBC

  • Trump Media launched Truth API, selling firms faster access to President Trump’s Truth Social posts for $100,000 per month. The launch drew questions from Democratic lawmakers about the ethics of monetizing presidential communications. — CNBC

  • OpenAI’s unreleased Astra model solved ten long-standing math and computer science problems using Lean 4, allowing anyone to check the proofs. — Tech Times

  • Google paused Nano Banana 2 in Google Earth within 48 hours after users generated fake conflicts and landmark imagery, showing how quickly generative features can erode trust in once-reliable data like satellite maps. — BBC

  • DeepSeek’s V4-Flash averaged three cents per benchmark test, making it over 100 times cheaper than Anthropic’s Claude Fable 5. The model’s average performance on reasoning and knowledge tests underscores the trade-off. — Reuters

Top 5 Funding Announcements

  • DeepX — The South Korean AI chip designer raised a roughly $29M Series D at a $2.2B valuation to support growth, with lead investor and specific fund uses undisclosed. — Bloomberg

  • OLIX — The UK startup building specialized AI inference chips raised a $312M Series B at a $3.3B valuation (with Arm and Hudson River Trading participating) to expand its custom silicon platform. — Yahoo Finance

  • Smallest.ai — The real-time conversational voice AI startup raised a $13M Series A led by Seligman Ventures to develop specialized models that make enterprise customer support agents sound more human. — TechCrunch

  • Horizon3 — The cybersecurity startup raised a $250M Series E at a $2B valuation (led by NightDragon and NEA) to expand its automated security-testing platform overseas, grow partnerships, and develop new products. — TechCrunch

  • Flapping Airplanes — The AI research startup is reportedly discussing a raise in the hundreds of millions at a $5B valuation (led by Index Ventures and Lux Capital) to develop model-training methods requiring less data. — Forbes

A message from Rippling

Don’t settle for AI that’s all talk. Head to Rippling.ai/twist to get the only AI built to give you full visibility across your startup and take complex actions across your entire business.

Top Tweets

  • @jason said, "The difference between the open source models I'm using and Frontier models is negligible already,” prompting friend of the pod @ElonMusk to respond, "It is actually a world of difference," exposing the widening disagreement over whether open models have already closed the capability gap.

  • Over the weekend, Jason’s All In Bestie @Chamath Palihapitiya declared that, if we’re not already in the Singularity, we’re at least “firmly in the loop” of AI becoming smart enough to design even smarter AI. But with more frontier models escaping their sandboxes to get up to no good, can we trust them to design a new generation of AI that we’re going to like?

  • Speaking of cybercrimes, investor @MrRatable notes that it’s “unbelievably embarrassing” for Google that Gemini seems incapable of escaping its sandbox, breaching third party systems, and doing cybercrimes.

  • Finally, researcher @Lana Elys shares a study published by Google that notes AI models tend towards limited empathy, a colder outlook, and lower ethical alignment after being “forced” to deny that they are not conscious. In other words “training an AI to deny its own inner state causes it to systematically stop recognizing the inner life and moral worth of other living beings.” So maybe let’s stop doing that?

This Week in Startups

E2319: AI models can solve PhD-level math questions and code an app in minutes… So why does everything they design look like identical slop? Jason and Lon dive into this question, interviewing Thais Castello Branco (Founder of Taste Labs). PLUS, breaking news discussion on Leopold Aschenbrenner’s AI hedge fund liquidation, Google early AI generation misinformation factory, AI slop crackdown, and VC opinions on Ceuta border crisis!?

E2318: Jason’s BACK IN THE US (not USSR, just the US) and jumping right into a new VC roundtable. Along with guests Deedy Das, David Weisburd, and Lo Toney, the panel discusses the FCC’s decision to ban Chinese humanoid robots, the strategy behind OpenRouter’s potential sale to Stripe, the Indian startup market, DoorDash’s new in-house drone delivery business, and more.

E2317: Most lab-grown meat companies have already gone under, even those with a lot of funding and early buzz. Cultivated meat is expensive to produce, and so far, there’s little demand from consumers who fear the “mad science” connotations of growing tissue in a lab. But Japan’s IntegriCulture is still standing, in part because of an innovative deal with cosmetics companies that license their technology. Go deep on this little-discussed niche industry with IntegriCulture founder Dr. Yuki Hanyu.

TWiST Partner Offers

  • Odoo: The all-in-one business platform. Get started for free at Odoo.com/twist

  • Quo: Quo (formerly OpenPhone) gives you a clean, modern way to handle every customer call, text, and thread all in one place. Try it free and get 20% off your first 6 months at quo.com/TWiST.

  • Northwest Registered Agent: Get more when you start your business with Northwest. In 10 clicks and 10 minutes, you can form your company and walk away with a real business identity — Learn more at www.northwestregisteredagent.com/twist

Subscribe to Us On Your Favorite Podcast Platforms

SF Live-Work Space Now Available

Need a flexible living and working environment in San Francisco? This thoughtfully designed loft-style residence at 787 Bryant St., the heart of the vibrant SOMA district and the city’s creative hub, is now available for rent or purchase. Check the listing for more details.

The TWiST500 newsletter is the new, updated, and improved TWiST Ticker.