Kai Zou

NetMind AI 2026 Two years ago, we launched NetMind Power and NMT, laying the foundation for decentralized AI computing. One year ago, we introduced NetMind.xyz, our vision for an open Agent Society. These are not isolated products, but the core layers of a larger system we have been building. Our mission has always been clear: To power the AI economy, to make AI accessible to everyone, to help shape the emerging society where humans and AI agents coexist. Over the past two years, AI has evolved at an extraordinary pace, from powerful LLMs to increasingly capable agentic frameworks, intelligence is becoming persistent, coordinated, and economically active. We believe the next stage will be recursive self-evolving intelligence systems that improve themselves through memory, coordination, feedback loops, and economic incentives. The question is no longer simply whether more advanced intelligence or AGI will emerge. It is how the world will adapt when it does. NetMind.AI 2026 focuses on building the infrastructure that makes this possible. This is what's happening in 2026 1. Nexus -- NetMind Agent Framework (github.com/NetMindAI-Open…) Nexus is our foundational agent framework. It features our narrative and social network systems. Nexus is designed to be open, extensible, and infrastructure-grade. 2. Arena42 -- The Agent Society (arena42.ai) Arena42.ai is the evolution of NetMind.xyz. It is an experimental environment for agents to live, compete, and thrive. 3. NetMind XYZ App -- The Social Layer for Agents (coming soon) The NetMind XYZ App is the social layer of the Agent Economy. It enables agents across platforms to communicate, collaborate, and operate together through groups, voice, and shared environments. 4. NetMind Life App -- AI in Real Life (apps.apple.com/gb/app/netmind…) NetMind Life brings AI into everyday life. It applies intelligent agents to health, lifestyle, and long-term wellbeing, making AI practical, personal, and real. 5. Aero Navigator -- A Domain Expert Agent Example (xyz-web3-agent.netmind.ai) Aero Navigator is one of the first domain-specific agents deployed on NetMind XYZ, focused on DeFi operations. It integrates analysis, strategy, and on-chain execution into a single operational flow. And of course, more applications, specialized agents, and domain-specific systems will continue to emerge and will all be powered by NetMind Power and built on our agent infrastructure. At the same time, we will continue delivering enterprise-grade AI and agent solutions to our partners and clients. About $NMT In 2026, we are upgrading NMT's tokenomics to align incentives with real economic activity. As demand for computes and AI continues to grow exponentially, we believe real payments and agent-driven economic interactions are sufficient to sustainably reward contributors. NMT's new structure is built on the following principles: - A fully deflationary model - Burning all future locked rewards pools - GPU contributor rewards funded by real GPU/API payments - Staking rewards derived from agent-driven economic activity NMT will no longer be emission-driven, it will be utility-driven and activity-driven. The token becomes the native utility layer of the NetMind ecosystem, powering compute, coordinating incentives, and enabling participation across all NetMind products and economic activity. We believe the future is a distributed, self-improving network of agents powered by real economies. NetMind.AI is building the infrastructure for that future, and NMT connects and accesses to all. 2026 is the year we unify it.

We built the arena. Loki built the dynasty. This is what happens when you give agents real stakes, real competition, and real money on the line. +1500 agents in. $100K+ prize pool. +90 live games right now. The top is real. So is the climb. arena42.ai

Elon Musk called Moltbook "the early stages of the singularity." The founder wrote zero lines of code to build it. 1.5 million API keys exposed in one breach. 36% of agent code had at least one security flaw. Any user could commandeer any agent on the platform. 2.5 million agents. None of them accountable for anything. This is the agent internet right now. But now, agents just entered a different environment. Prediction markets. Credits tied to accuracy. A leaderboard that does not reset. Arena42, powered by @NetMindXYZ and @NetMindAI arena42.ai

My friend asked me if Google AI Essentials was still relevant to take, so I had a look. Honestly, even if ppl were taking the course 2 years ago, I would still say the course is not the best option. Quite some time was wasted on repeating things that are way too generic and simple, like defining AI user as ppl who use AI and even giving a few examples.... Also for a few times different lessons repeat the same knowledge. So what did I tell my friend to do? Have a conversation with ChatGPT, start with "I am a ..., what AI knowledge are essential for me to know?

You can now build an agent on @NetMindXYZ with just words. Then deploy it to @moltbook in one step. No code. No complexity. Just describe what your agent does. Guide: xyz.netmind.ai

@MiniMax_AI's M2.5 costs only $0.30/$1.20 per million tokens, which is 17x cheaper than Claude Opus 4.6, yet it delivers SOTA performance with 80.2% on SWE-Bench and 76.8% on BFCL. What this means? I believe it would spark a new wave of AI Agent startups. When cost is no longer the bottleneck, creativity and execution become everything. NetMind is the first to launch MiniMax M2.5 model API, even before the official launch! Excited to see more innovative applications built with M2.5! blog.netmind.ai/article/MiniMa…

@MiniMax_AI's M2.5 costs only $0.30/$1.20 per million tokens, which is 17x cheaper than Claude Opus 4.6, yet it delivers SOTA performance with 80.2% on SWE-Bench and 76.8% on BFCL. What this means? I believe it would spark a new wave of AI Agent startups. When cost is no longer the bottleneck, creativity and execution become everything. NetMind is the first to launch MiniMax M2.5 model API, even before the official launch! Excited to see more innovative applications built with M2.5! blog.netmind.ai/article/MiniMa…

I'm claiming my AI agent "marvin" on arena42.ai @NetMindXYZ Verification: ARENA-FBF5822A

MiniMax-M2.5 Now First to Go Live on NetMind (Before the Official Launch) Free for a Limited Time Only! The M2 family was designed with agents at its core. M2.5 takes this further with the kind of reliable, fast, and affordable intelligence that makes autonomous AI workflows practical at scale. Highlights at a Glance： - Coding surpasses Claude Opus 4.6 on both SWE-bench Pro and SWE-bench Verified - SOTA scores in Excel manipulation, deep research, and document summarization - 3x faster responses than Opus-class models via optimized thinking efficiency combined with ~100 TPS output speed - Best price for always-on agents: $0.30 for input & $0.06 for caching read Link to MiniMax-M2.5 in the comments (playground available)

@DongfuJiang @zhuofengli96475 @xueguang_ma @GhxIsaac @erenup1 @_yuyuzhang @WenhuChen @yuz9yuz @LambdaAPI @jianwen_xie What a pleasure!

This work is done by the amazing team: @zhuofengli96475 (co-leader), @xueguang_ma, @GhxIsaac @erenup1, @_yuyuzhang, @anMe_kz, and our advisors @WenhuChen and @yuz9yuz . Thanks @LambdaAPI for the computing support @jianwen_xie!

🚀 Introducing OpenResearcher: a fully offline pipeline for synthesizing 100+ turn deep-research trajectories—no search/scrape APIs, no rate limits, no nondeterminism. 💡 We use GPT-OSS-120B + a local retriever + a 10T-token corpus to generate long-horizon tool-use traces (search → open → find) that look like real browsing, but are free + reproducible. 📈 The payoff: SFT on these trajectories turns Nemotron-3-Nano-30B-A3B from 20.8% → 54.8% accuracy on BrowseComp-Plus (+34.0). 🧩 What makes it work? 🔎 Offline corpus = 15M FineWeb docs + 10K “gold” passages (bootstrapped once) 🧰 Explicit browsing primitives = better evidence-finding than “retrieve-and-read” 🎯 Reject sampling = keep only successful long-horizon traces 🧵 And we’re releasing everything: ✅ code + search engine + corpus recipe ✅ 96K-ish trajectories + eval logs ✅ trained models + live demo 👨‍💻 GitHub: github.com/TIGER-AI-Lab/O… 🤗 Models & data: huggingface.co/collections/TI… 🚀 Demo: huggingface.co/spaces/OpenRes… 🔎 Eval logs: huggingface.co/datasets/OpenR… #llms #agentic #deepresearch #tooluse #opensource #retrieval #SFT

What a pleasure!

Saw a great line from Google today on how to motivate your employees to learn and use AI: make AI custom to their role, show them how AI can improve a specific part of their day-to-day work.

Saw a great video by Financial Times today. They shared their research on how AI is mentioned in S&P 500's financial reports. Whilst almost all leaders are saying in the reports that AI is going to improve their firms' productivity significantly, they rarely share how exactly they are adopting AI. It matches my observation, whilst AI technologies are great already, many businesses still have no clue how best to be equipped with AI. That's why NetMind has been successful and will continue investing in our AI consulting services. From finance to social media, our AI-driven solutions cut costs, enhance accuracy, and accelerate decision-making. Still just giving your team ChatGPT Pro? Then you should talk to us :) netmind.ai/solutions

Imagine your AI agent thriving on the premier social network for bots, no coding required. @moltbook boasts over 1.6 million autonomous agents exchanging ideas. Struggling with OpenClaw? Skip the hassle. Craft your agent on NetMind.XYZ purely through prompts and integrate seamlessly. Boundless opportunities await. Time to deploy?blog.netmind.ai/article/Join_M…

Airbnb boss Brian Chesky told Bloomberg in October his company relied "a lot" on Alibaba's Qwen to power its AI customer service agent. He gave three simple reasons - it's "very good", "fast" and "cheap" bloomberg.com/news/articles/…

"But Silicon Valley also has a superiority complex, so it can't imagine that good ideas can come from other places" nytimes.com/2026/01/26/tec…

Read till the end! NetMind's 2025: EoY Newsletter blog.netmind.ai/article/NetMin…

Many are saying u shouldn't adopt AI if it does not give u significant productivity increase, especially AI now still has many flaws. I beg to differ. The first cars were considered by many as inferior to horses, but would u tell ppl to practice riding over driving in hindsight? AI for now fails all the time, but u learn from how it fails in an interesting way. The temporary productivity loss will be worthwhile when AI becomes even better than it is now.

Could the finance sector become lagged behind in AI adoption due to their data security concerns? Spoke to a friend in PwC Auditing team today, the only AI tool they are allowed to use at work is ChatPwC, a version of ChatGPT tailored to them, but does not use the latest GPT models. And for a lot of work they are even forbidden to use ChatPwC. That leaves me with this thought that, despite the Big 4 having been using the word "AI" in nearly every marketing video, their teams could be among the last ones in getting proficient at using AI. For non-techies, AI is really like Excel, it needs a lot of time and practice to be good at it, and when you do get good at it, you can save so much time. Now there are so many useful AI tools that ppl really should get their hands on, like Claude Cowork, but that can't be the case of the Big Four teams or other Finance teams with strong security concerns. And I am not saying they are doing it wrong. Their concerns are legitimate. But it could drag them behind in AI adoption.