In partnership with
In Today’s Issue:
🎨 Alibaba’s Qwen3-Max-Thinking and Moonshot’s Kimi K2.5 launch with "Swarm Intelligence”
🛰️ The world’s first fully open AI weather stack launches
🔋 The new Tianxing II battery enters mass production
⚠️ Anthropic’s CEO warns at Davos that superhuman AI could arrive by 2026
✨ And more AI goodness…
Dear Readers,
Today's AI landscape just shifted, again. While the West debates regulation, China dropped two paradigm-breaking models in 48 hours: Alibaba's Qwen3-Max-Thinking, which autonomously chooses its own tools mid-conversation, and Moonshot's Kimi K2.5, capable of spawning 100 specialized AI agents to tackle complex tasks in parallel.
Meanwhile, NVIDIA democratized weather forecasting with Earth-2, cutting compute costs by 90% and shrinking 15-day global predictions from hours to seconds. Microsoft taught robots to feel with Rho-alpha, fusing vision, language, and tactile sensing into bimanual systems that learn from human corrections in real time. Add CATL's sodium-ion batteries promising 3-million-kilometer lifespans at a quarter of lithium's cost, and Dario Amodei's stark warning that superhuman AI could arrive within two years, and you've got a week that redefined what's possible, and what's at stake. Let's dive in.
All the best,
🌍 NVIDIA Opens AI Weather Future
NVIDIA has launched the Earth-2 family, the world’s first fully open, GPU-accelerated AI software stack for weather and climate forecasting, spanning everything from real-time storm nowcasting to 15-day global forecasts. The open models cut compute time from hours to seconds, outperform traditional physics-based systems in key benchmarks, and are already delivering up to 90% compute savings for national weather agencies and energy operators. This move democratizes high-precision weather intelligence, unlocking faster decisions in disaster response, energy grids, finance, and climate risk worldwide.
🤖 Humanity Faces AI Adolescence Test
In his latest essay, Dario Amodei argues humanity is entering a decisive “technological adolescence,” where superhuman AI could arrive within 1–2 years and reshape security, biology, geopolitics, and work itself. He warns of five major risks, runaway AI autonomy, bioterror misuse, AI-enabled authoritarianism, mass job displacement, and unpredictable second-order effects, while rejecting both doomerism and complacency. Amodei calls for urgent but surgical action: better AI alignment, transparency laws, chip export controls, bioweapon guardrails, and economic reforms to ensure AI’s massive upside doesn’t fracture society.
🔋 CATL Launches Sodium Battery Breakthrough
CATL has unveiled the world’s first mass-produced sodium-ion batteries for commercial vehicles, delivering up to 10,000+ charge cycles, strong cold-weather performance (90% capacity at –40°C), and costs as low as $10–19/kWh (4x cheaper than lithium-ion). The Tianxing II lineup targets vans and trucks, with sodium versions rivaling LFP energy density (175 Wh/kg) while potentially lasting 3 million km, plus lithium variants offering 18-minute fast charging and 800 km range for logistics. With a planned European factory by 2026 and sodium batteries poised to cover 40% of China’s EV demand, this tech could dramatically cut EV costs and accelerate global adoption, despite sodium’s heavier weight tradeoff.
OpenAI Town Hall with Sam Altman
Swarm Intelligence Meets Reasoning Models
The Takeaway
👉 Qwen3-Max-Thinking introduces adaptive tool-use: The model autonomously decides when to search, run code, or use memory—eliminating manual tool selection and streamlining complex workflows.
👉 Kimi K2.5's Agent Swarm enables parallel execution: By spawning up to 100 specialized sub-agents, complex tasks that previously took hours can now complete up to 4.5x faster through coordinated parallel processing.
👉 Both models are fully open-source: Developers worldwide can download, customize, and deploy these trillion-parameter models without licensing costs, accelerating adoption across startups and enterprises.
👉 Chinese open-source AI now commands ~30% of global downloads: This momentum signals a fundamental shift in where frontier AI development originates—and who controls its distribution.
The AI world just witnessed another seismic double-punch from China. Within days of each other, Alibaba's Qwen team and Moonshot AI dropped two powerhouse models that have developers and researchers buzzing with excitement.
Qwen3-Max-Thinking, Alibaba's latest flagship, brings a fascinating twist: adaptive tool-use capabilities that autonomously select and leverage built-in Search, Memory, and Code Interpreter capabilities during conversations, and an insane eval of 58.3% in HLE with tool search. This means the model decides when to reach for external tools without you having to tell it.
Meanwhile, Moonshot AI unleashed Kimi K2.5 with something even more radical: an "Agent Swarm" system. For complex tasks, Kimi K2.5 can self-direct an agent swarm with up to 100 sub-agents, executing parallel workflows across up to 1,500 tool calls. Imagine an AI that spawns its own specialized team members on the fly.
Both models arrive fully open-source, continuing China's aggressive commitment to accessible AI. Chinese companies have been rolling out open-source models in rapid succession, enabling the development of AI systems for a wide range of applications at lower cost.
What makes these releases special isn't just raw performance, it's the architectural creativity. From experience-cumulative reasoning to swarm intelligence, Chinese labs are pushing boundaries in ways that challenge conventional thinking.
Why it matters: These releases demonstrate that frontier AI innovation is now truly global. The open-source approach democratizes access to cutting-edge capabilities, potentially reshaping how developers and businesses build AI-powered products worldwide.
Sources:
🔗 https://qwen.ai/blog?id=qwen3-max-thinking
🔗 https://huggingface.co/moonshotai/Kimi-K2.5
AI in HR? It’s happening now.
Deel's free 2026 trends report cuts through all the hype and lays out what HR teams can really expect in 2026. You’ll learn about the shifts happening now, the skill gaps you can't ignore, and resilience strategies that aren't just buzzwords. Plus you’ll get a practical toolkit that helps you implement it all without another costly and time-consuming transformation project.
Microsoft's Rho-alpha: Robots That Feel
Imagine a robot that doesn't just see and hear your commands but can actually feel its way through complex tasks, adjusting on the fly when it encounters resistance, learning from your corrections, and adapting to the messy, unpredictable real world we all inhabit. Microsoft just made that vision tangible with Rho-alpha, their first robotics model derived from the Phi series that translates natural language commands into control signals for bimanual manipulation tasks. What makes Rho-alpha special isn't just another incremental improvement, it's the fusion of vision, language, and tactile sensing into what Microsoft calls a "VLA+" model, adding a sense of touch that previous robotics AI simply lacked.
Announced on January 21, 2026, Rho-alpha promises bimanual dexterity that past architectures rarely achieved outside controlled labs. The model can execute commands like "insert the plug into the bottom socket" or "pull out the red wire" with impressive precision, and when it struggles, human operators can intervene with 3D input devices to provide corrective feedback that the system learns from in real time. Microsoft is already testing Rho-alpha on dual-arm setups and humanoid robots, with plans to make it available through their Research Early Access Program and eventually via Microsoft Foundry.
Microsoft Research blended three data sources: human teleoperation trajectories on real hardware, synthetic demonstrations generated with NVIDIA Isaac Sim running on Azure clusters, and web-scale visual question-answering images to strengthen semantic grounding. This hybrid approach solves one of robotics' biggest challenges: the scarcity of diverse training data, especially for tactile feedback. By combining simulation with real-world demonstrations, Rho-alpha can generalize to new situations without requiring exhaustive physical testing for every scenario.
Equipment policies break when you hire globally
Deel’s latest policy template on IT Equipment Policies can help HR teams stay organized when handling requests across time zones (and even languages). This free template gives you:
Clear provisioning rules across all countries
Security protocols that prevent compliance gaps
Return processes that actually work remotely
This free equipment provisioning policy will enable you to adjust to any state or country you hire from instead of producing a new policy every time. That means less complexity and more time for greater priorities.
