New top AI model Kimi K3 from China, Google Gemini falters, Apple revs up on AI & More. AI-RTZ #1151

1. Anthropic & OpenAI now up against Kimi K3 from China: Moonshot AI took the world’s open source frontier AI model crown this week with the release of the world’s largest open source, 2.8 trillion parameter frontier class AI model, Kimi K3. China now has the dibs on the world’s largest open source AI model, just in time for its World AI conference this week, where President Xi delivered a keynote highlighting China’s AI leadership. As I discussed on ARD #120, Kimi K3 lands close to Anthropic’s Opus 4.8 — but crucially it’s open-weights (the world can inspect, host and run it independent of China) at perhaps a third of the price or less. By Friday’s ARD #121 the point got sharper: Kimi K3 has effectively upset the global lead the two closed frontier labs assumed — Anthropic and OpenAI would likely have been down 10%+ on the week if they were already public (not stock advice). My take: this is less a binary “race” than a global undertaking, where US firms increasingly lean into Chinese open models as portfolio supplements — balancing cost, privacy and capability — even as geopolitics tries to pull the two apart. The perceived US–China gap keeps looking narrower than assumed. More here and here.

   

2. Google Gemini now behind frontier AI leaders: Google Gemini 3.5 Pro now is lagging the US Frontier AI model companies Anthropic and OpenAI. It’s flagship Gemini 3.5 Pro is now behind in schedule by months. As I covered on ARD #121, the slip lands specifically in the AI coding race — the agents/coding/productivity layer where Anthropic’s Fable 5 and OpenAI’s GPT-5.6 are surging — with Gemini 3.5 Pro reportedly held back for falling short of internal goals. It’s the latest twist in Google’s whipsaw Gemini arc: a “Code Red” when ChatGPT first launched, then a resurgent Gemini that handed the “Code Red” back to OpenAI, and now another turn back. My take: this is a stumble, not a structural break — Google still has a similar opportunity to turn it around, it’ll just take more time, and it remains among the best-positioned globally on consumer AI, side by side with Apple. More here.

   

3. Apple Revs up in AI: Apple had a big week. Apple had a big week. First a trade secrets lawsuit against OpenAI on AI Devices, then an updated AI chip roadmap for Apple Intelligence on local devices and servers. And finally, some partnerships with Alibaba and Baidu to use their open source AI and Search technologies in upcoming Apple Intelligence and Siri AI driven iPhones in China. The lawsuit — Apple’s “First Blood” against OpenAI over AI-device trade secrets (400+ hires under Jony Ive) — lands pointedly ahead of OpenAI’s expected mega-IPO, as I covered around ARD #117. The chip roadmap (M6, M7 Pro/Max/Ultra into M8, touch MacBook Pro) underlines Apple’s under-appreciated silicon strength for running Apple Intelligence locally and on its privacy-protected servers — the Srouji/Ternus thread from ARD #118. And by Friday’s ARD #121, the China approval with Alibaba (Qwen) and Baidu — both stocks jumping in Hong Kong — showed Apple navigating US–China geopolitics through local partners, exactly as it uses Google Gemini and Nvidia in the US. My take: Apple keeps executing on its AI opportunities with alacrity, worldwide. More here, here and here.

   

4. Nvidia backed Thinking Machine’s US open source AI model: Thinking Machines founded only three years ago by ex-OpenAI senior talent, launched its first open source AI model Inkling this week. Nvidia is a key investor and backer, providing the company with lots of its latest Vera Rubin chips to train and develop the model. It’s part of Nvidia’s ‘Kingmaker‘ strategy in US open source AI companies. The last one being Reflection AI that I’ve discussed. As I noted on ARD #120, Inkling — from the team led by ex-OpenAI CTO Mira Murati — isn’t trying to go toe-to-toe with the frontier flagships; it’s a flexible, quirky open-weight tool aimed at builders. The bigger signal is the pattern: Nvidia increasingly playing kingmaker, seeding a stable of US open-source champions — Reflection AI, now Thinking Machines — with capital and its latest Vera Rubin silicon. My take: it’s the US answer to China’s open-source surge (Kimi K3, Qwen, DeepSeek), and it fits the argument I’ve made that Nvidia and Apple can anchor a US open-source counterweight rather than cede that ground. More here.

   

5. China’s DeepSeek back for Seconds: China’s DeepSeek is potentially back for its second major funding round at a higher $70+ billion valuation, a month after raising billions already. As I flagged on ARD #119, DeepSeek “coming back for seconds” so soon — at a markedly higher valuation — is another data point in the story running through this whole week: China’s open-model players (DeepSeek, Moonshot with Kimi K3, Alibaba’s Qwen) are all coming on strong, and Asian markets have been hitting new highs alongside them. My take: it reinforces that the US–China long-term AI opportunity is getting bigger even as geopolitics tries to shrink it — money is flowing into Chinese AI at valuations that assume the gap with the US frontier keeps narrowing. More here.

   

Other AI Readings for weekend:

1. EU rules against Google on DMA, following earlier tussles with Apple and other US big tech. More here.

2. Taiwan Semiconductor, TSMC, ‘the Fed of Global Tech/AI’, expands global chip fabrication capacity. More here.

(Additional Note: AI Ramblings Daily on YouTube, is now a weekday Daily podcast called AI Ramblings Daily (ARD). Just passed the #100 podcast milestone mark. Different content than AI-Reset to Zero (AI-RTZ) substack, which remains a daily morning substack write-up. With now over 1140+ ‘MY TAKES’ on key AI events and issues turbulently flowing by. ARD is typically a 20 minute afternoon podcast every day, on my take on additional AI developments. Both daily substack and podcasts typically discuss different AI issues and items. There is a daily text summary of the daily podcasts here on the substack, as well as one minute YouTube ‘Shorts’ video clips on the key topics discussed. And all are free to subscribe for now. Try this week’s series with ARD Episodes 117 — Too Much of a Bad Thing, 118 — Shift Happens to IBM, Meta & Microsoft, 119 — Get ‘While the Getting Is Good’, 120 — Drumbeats of AI Progress March On, and 121 — The Many AI Twists and Turns Accelerate here):

Up next, the Sunday ‘The Bigger Picture’ tomorrow. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)