Meituan Drops the Hammer: LongCat-2.0 Open Source & 1.6 Trillion Parameters
Mark your calendars for June 30th—Meituan just pulled back the curtain on their brand new heavyweight, LongCat-2.0, and they’re making it completely open source. This isn’t just another model; it’s a beast trained entirely on a domestic cluster of 50,000 GPUs. We’re talking about a 1.6 trillion parameter giant (with an average activation of roughly 48B and a dynamic range between 33B and 56B). It’s been fed over 30T tokens of data covering Chinese, English, multilingual text, and code, all while natively handling context windows as long as 1 million tokens.
NVIDIA’s Robot Squad is Hiring Big Time in Beijing, Shanghai, and Shenzhen
According to CCTV Finance, back on June 29th, NVIDIA announced they are officially opening their doors for talent in the robotics space. The team is looking for pros across four key pillars: embodied intelligence, simulation, deployment, and solution architecture. And guess where? They’ve set up shop in Beijing, Shanghai, and Shenzhen.
Here’s the deal from the team: Their Embodied Intelligence group is diving deep into dexterous manipulation, wearable sensor modeling, and full-body control to build the next generation of general-purpose robots. The Simulation crew is busy building the virtual training grounds so robots can learn faster and more reliably before hitting the real world. Then there’s the Deployment team, focused on optimizing algorithms for humanoid robots and getting them off the screen and onto the floor. Finally, the Solutions team is taking these cutting-edge techs straight into industrial and service sectors.
Huawei Goes All-In: OpenPangu-2.0-Flash Model Live Now
Another big day on June 30th for Huawei! They’ve officially thrown open the gates for the OpenPangu-2.0-Flash model, packing a punch with 92 billion parameters. If you’re ready to get your hands dirty, the weights, basic inference code, and training/inference operators are live on their open-source platform right now. Keep an eye out, because the heavier-duty OpenPangu-2.0-Pro version (weights and code) drops in July, with even more components rolling out later this year.
Kimi Valuation Skyrockets to $31.5 Billion: Is It the Next Anthropic?
Get this: Moonshot AI’s Kimi just closed a massive round at a $20 billion pre-money valuation, but that was yesterday. Today, we’re hearing whispers of a new round pushing the pre-money valuation all the way up to $31.5 billion. Insiders close to Kimi revealed some juicy numbers during the talks: by mid-June, their Annual Recurring Revenue (ARR) smashed through the $300 million mark.
So, what’s driving this explosion? It’s simple: better models are attracting more developers and spiking API usage. In fact, API income now makes up over 70% of their total revenue and keeps climbing. The growth curve looks eerily familiar to Anthropic’s early days—massive developer adoption, skyrocketing API reliance, a surge in paying international users, and price hikes driven by raw model capability.
Xiaopeng Motors’ Robot Chief Miliangchuan Has Stepped Down
Breaking news from June 30th via 21st Century Business Herald:Miliangchuan, the head of Xiaopeng Motors’ robotics division, has recently left the building. To fill the void, CEO He Xiaopeng is now personally wearing two hats, serving as the interim head of both the Xiaopeng Robotics Center and its Product Department.
UBTECH Reveals Prices for Full-Sized Ultra-Humanoid Robots
Also on June 30th, UBTECH dropped the official pricing for their latest star, the U1 full-sized ultra-humanoid robot. With 88 degrees of freedom, this thing is impressive. Here’s the damage: The U1 Pro comes in at 169,800 RMB. The U1 Ultra versions are priced at 990,000 RMB for the male variant and 880,000 RMB for the female. If you want something lighter, the semi-body U1 Lite is available for 119,800 RMB.
FactSet Teams Up with Google Cloud to Build Financial AI Agents
On June 30th, global financial data powerhouse FactSet announced a major strategic partnership with Google Cloud. They’re joining forces to build next-gen AI agents powered by the Gemini model. The collaboration zeroes in on three game-changers:
First, they’re embedding Google Enterprise Search and the Gemini model directly into FactSet workstations to supercharge research depth using Google’s reimagined features. Second, they’re deepening the integration between FactSet and Google Cloud’s Gemini Enterprise to ensure seamless interoperability between the workstation and the cloud platform. Third, they’re co-developing intelligent agent workflows to revolutionize portfolio management, trading advisory, and corporate finance. Plus, FactSet is adding Google Cloud to its infrastructure mix to optimize their backend.
