Wishing you a bright and peaceful Mid-Autumn Festival. 🌕 As the full moon brings people together, AiWork hopes to make work feel a little more connected too — helping teams turn ideas into actions, simplify daily tasks, and create more time for what truly matters. May this season bring you reunion, clarity, and new momentum.

AiWork 全新升级了！🎉 把 AI 真正放进了日常工作流里：写方案、做分析、生成文档、研究报告、代码开发、任务协同……很多过去需要来回切换工具、反复沟通推进的事，现在都可以交给 AiWork 跑起来。 AiWork·Code: 从一个需求开始，能一路推进到可验收交付，不只是辅助写代码，而是在补齐工程里的规划、执行、检查和交付能力。 AiWork·Work: 像一个日常办公助手，把资料整理、内容生成、任务协作这些琐碎但耗时间的事接住，让人可以把精力放回判断和决策上。 把繁琐交给 AI，把时间留给你！ aiwork.dtagi.cn

Just accepted to VLDB 2027! We’ve partnered with Tsinghua University (@Tsinghua_Uni) to release AgenticDataBench: a Comprehensive Benchmark for Data Agents. Why does this matter? Because evaluating AI agents shouldn't be a guessing game. We built it on 3 core elements: ✅ Real-world business tasks ✅ Cross-domain data ✅ Fine-grained skill analysis 💡 For enterprises: A scenario-driven baseline to compare agent capabilities. 💡 For devs: Clear diagnostics to fix shortcomings and align with actual production needs. Fully open-source. Ready to test your data agents today? 👇 🔗 agenticdatabench.github.io #DataAgents #AI #VLDB2027 #AgenticDataBench

以前用工具，更多是在“告诉它下一步怎么做”；但 AiWork 不太一样，你说清楚目标之后，自己去拆任务、调工具、跑流程、遇到问题再修正，最后把结果交回来。 一句话可以被拆成很多连续动作，比如理解需求、查资料、写文件、调用工具、做检查，甚至多个 Agent 一起协作。 这才是 AiWork 真正有意思的地方——它把复杂工作变成一条能跑通的任务链。

一个想法到交付，中间到底经历了什么？ AiWork Space：从云端 Agent、MCP、Skill，到 CLI 和 Workflow，每个环节都不是单点能力，而是在帮团队把研发流程跑顺。 好的工具不只是让个人写得更快，而是让团队交付更稳。当环境、规范、流程都能沉淀下来，研发就不再靠“口口相传”和“个人经验”推进，而是有一条清晰、可复用、可持续优化的路径。 从一个想法，到真正可交付，Space 想解决的就是这中间最复杂、也最容易被忽略的部分。

Just had AiWork set up a recurring zkVM research digest. Twice a week, it scans papers, conferences, and key repos, skips the PR noise and repeats, and sends only the updates that matter—with sources, risks, and follow-ups. Set it once, stay up to date. aiwork.dtagi.ai/en/home #AiWork #AIAgent #AICoding

zkDTVM has become the first zkVM on Ethproofs (@eth_proofs) to achieve real-time proving using just 4 RTX 5090 GPUs. 99% of proofs <10s with an average proving time of 4.6s. Proof size remains under 300 KB while maintaining 128-bit security. zkDTVM will be open-sourced soon. We look forward to the community embracing and leveraging it.

AiWork 凭什么能自主把活干完？ 在 AiWork Meetup 上，我们揭秘了 Agent 的核心逻辑： ✅ 先查事实：拒绝盲目执行，基于真实数据行动。 ✅ 只问关键：减少无效沟通，直击痛点。 ✅ 生成计划：从需求拆分到测试、提PR，全流程自动化！ 不止于个人提效，更是打破团队协作壁垒 ，通过 Workflow 统一团队标准流程。 让 AI 真正干活，实现可靠交付。智能工作，从此简单！ #AiWork #AIAgent #Workflow

以前总以为“准点下班”是奢望，直到把任务甩给了 AiWork Space。 它不是工具，它是那个永远在线、从不抱怨的 24 小时 AI 同事。 ✅ 睡前丢任务，醒来直接验收 ✅ 临时插单？不存在的焦虑 ✅ 全流程自动化闭环 ✅ 不懂技术也能提需求 这次，真的可以准点下班了。 你的云端同事，已就位。 aiwork.dtagi.cn/agent #AiWork #AI同事 #云端Agent

接手五年老项目，代码多看一眼就想辞职？ 别慌！AiWork 专治“老项目迷路症”。 它不是一上来就重构代码，而是先生成一张项目导航图。 ✅ 四步盘清仓库 ✅ 节点来自真实目录，依赖来自真实导入关系 ✅ 只回答四件事：谁该读、怎么走到代码、哪些是非目标、验证入口在哪 每个分区配 Playbook，人和 Agent 看同一张图，最后直接生成一个本地可打开的 HTML 导航图，先把 AI 友好的地形修好，再让它开工！

为什么你的 AI Agent 总是“翻车”？ 不是模型不够聪明，而是它不够“稳”！ AiWork 算法工程师分享实战经验： 🔹 Terminal Bench 2.1：全球第 6（78.7%） 🔹 AgenticDataBench：dev 榜单第 2、test 榜单第 1 🔹 用更便宜的 Kimi K2.5 追平 Opus4.8 #AiWork #Opus #Kimi

AiWork是什么？ 我们来听听运营、前端、后端研发、产品经理说些什么。

AiWork 首场技术 Meetup，见证自研引擎的力量！ 🚀 我们将首次公开亮相 AiWork 自研核心引擎，并毫无保留地分享一线最佳实践案例。从底层架构到工程落地，带你深入理解下一代 AI Coding 工具的核心逻辑。 📅 时间：2026 年 9 月 12 日 📍 地点：上海 · 蚂蚁 S 空间 👇 席位有限，立即报名锁定名额： aiwork.dtagi.cn/docs/activit… #AiWork #技术Meetup #AI编程 #上海活动

Statistics from a 3k block sample.

Our zkDTVM v0.8.0 8-GPU prover is now live on Ethproofs (@eth_proofs). 🔹 99% proving < 6.1s 🔹 Avg proving time: 3.0s 🔹 Security: 128-bit (UDR+JBR) 🔹 Proof size < 300 KB Fully meets the Ethereum RTP standard！ #ZK #zkEVM #EthProofs #ETH

AiWork 重磅上新，三款强力模型已就位！满足多样化需求，欢迎前往体验~ ⚡ ling-3.0-flash-fin【限时免费】🔥🔥🔥 • 金融领域专精模型，懂业务更懂风控。 🧠 qwen3.8-max • 综合能力较强，复杂任务交给它。 🎨 gpt-image-2 • 高质量图像生成，内容创作轻松搞定。 中文站：aiwork.dtagi.cn 国际站：aiwork.dtagi.ai

Ling-3.0-flash-fin is now available in AiWork — and it’s FREE for a limited time! As the first finance-enhanced model in the Ant Ling family, Ling-3.0-flash-Fin was developed by Ant Group alongside leading financial institutions and domain experts. It builds on Ling-3.0-flash with continued training on high-quality financial data. #AiWork #AntGroup #Ling3 #FinTech #AI

Same model. Same tasks. Same grader. We changed only the coding harness. AiWork shipped 31% more delivered code from about a third of the output tokens, in 45% less time — with every app passing. The model sets the ceiling. The harness decides how much of it you keep. 🧵 🔗 openlabs-intl.antdigital.com…

Got an image you want AI to break down for you? No upload. No drag-and-drop. Just Ctrl + V — paste it in, and AiWork gets it instantly. Screenshots, reports, architecture diagrams, error logs… And yes — even the @eth_proofs prover leaderboard. Paste it. Ask. Done. 👉 aiwork.dtagi.ai/en/home

🚀 AiWork Space 正式上线！ 您的 7×24 小时云端 AI 专家团，随时待命 云端一站式搭建，开箱即用 内置 AI 专家团，覆盖研发全流程 最佳实践工作流，拿来就能跑 👉 aiwork.dtagi.cn/agent

The biggest time sink in app development isn't writing code. It's watching your Agent loop back to the same conclusion — again and again. So we put it to the test. Same model. PRDBench. 5 PRD apps. ✅ AiWork.Code → 2h 26m ❌ Competitor baseline → 4h 28m 45% faster. Not a marginal gain — two full hours back, with a more complete build. Less waiting. More shipping. ➡️ aiwork.dtagi.ai/en/home

The hardest part of code security: separating real exploits from 1,000s of alerts. AiWork found an unauthenticated RCE in a major distributed scheduler’s default setup, traced the full attack chain, and delivered verified proof. 5,369 vulns found | 302 critical | 1,653 high. Learn more 🔗 openlabs-intl.antdigital.com… We leave code security to AiWork.

感谢大家一路支持 DTCoder，反馈量真的超出想象。🙏 接下来，我们将解锁 CoWork 能力——让 AI 从“写代码”进化到“协同工作”。 为此，我们正式宣布：DTCoder 升级为 AiWork。 AiWork：智能工作，从此简单。 名字变了，初心不变：让每一次交付，都生产级就绪。🚀 🔗 国际站点：dtcoder.ai/en/home 🔗 中国站点：dtcoder.antdigital.com/home

🔥 Following the enthusiastic feedback around DTCoder, we are about to launch the work capability — so we are officially upgrading our AI tool name to AiWork. AiWork: Smart work, made simple. 🔗 dtcoder.ai/en/home 🔗 China Site: dtcoder.antdigital.com/home #AI #AICoding #AICowork

DTCoder：让 AI 从「能写代码」升级为「按工程纪律可靠交付」的 Coding 工具。 免费领取您的 Credits：dtcoder.antdigital.com/ #Agenticcoding #AI #Agent

Today, we are announcing 18 million devices on-chain! On our journey to bridge the real world and Web3, we’ve just taken another giant leap forward! Large-scale, high-value trusted physical assets are now successfully on-chain! We are breaking the boundaries of traditional finance, bringing unprecedented liquidity and vitality to physical assets on-chain! 18 million physical devices are "breathing" on-chain in real time! The integration of massive IoT devices injects the most authentic off-chain data into on-chain assets, creating an absolutely trustworthy on-chain ecosystem! 18 Million Devices = A Future of Infinite Possibilities! Thank you to every ecosystem builder for your support. We will keep accelerating to ensure real-world value flows freely on-chain! Let’s witness the stars and seas of Web3 together!

How to run loop engineering across the full development lifecycle? We propose a three-layer agentic workflow framework centered on the critical challenge of rule codification: 🔹 Unify the general development process to ensure alignment and baseline execution across all projects. 🔹Distill project rules from Agent coding sessions, leveraging hook mechanisms as the binding agent. 🔹 Close the loop spanning Process → Artifacts → Rules → Automation. openlabs-intl.antdigital.com… #AICoding #LoopEngineering

When AI Coding Goes Deep into the Development Process, the Hardest Part Isn't Writing Code. This post provides a brief overview of the real-world AI coding engineering practices adopted by engineers at AntChain OpenLabs. 🔗 openlabs-intl.antdigital.com… Chinese 🔗: mp.weixin.qq.com/s/N3BYiZhIq…

Our zkDTVM 8-GPU prover is now live on Ethproofs (@eth_proofs). zkDTVM is the world's first zkVM to achieve real-time proving on 8x 5090s: 🔹 99% proving < 7.7s 🔹 Avg proving time: 4.64s 🔹 Security: 100-bit (UDR) 🔹 Proof size < 600 KB Next: Proof size < 300 KB & 128-bit security. #ZK #zkVM #Ethereum #EthProofs #RealTimeProving

Post-quantum signatures are orders of magnitude larger than ECDSA — a dealbreaker for high-throughput blockchains. Our zkDTVM changes that: 4,500 leanXMSS sigs (~5.3 MiB) is compressed into one ZK proof (~577 KiB). Single GPU results: ⚡ 17,857 leanXMSS verifications/sec ⚡ 21× faster than leanVM (CPU prover) ⚡ End-to-end quantum-resistant Next up: SPHINCS+. Full writeup 👇 openlabs-intl.antdigital.com…

Last week, zkDTVM proved 100% of Ethereum blocks. Every proof was under 8.8s, with an average time of just 3.1s. Fastest. Most Stable. Most Reliable.

Another record broken! With better multi-GPU optimizations, zkDTVM’s avg proving time on 16 GPUs is now 3.5s. #zkDTVM #EthProofs #Ethereum #GPU #ZK

zkDTVM enters the RTP cohort on @eth_proofs: 99.91% liveness, 99.91% of proofs < 10s.

Ant Digital Technologies' zkDTVM Tops Benchmark "EthProofs" In the recent global benchmark test "EthProofs" initiated by Ethereum, Ant Digital Technologies' Zero-Knowledge Proof Virtual Machine (zkDTVM) achieved a new record - block proving time of 3.7 seconds, securing the No. 1 spot globally. This breakthrough also provides new capabilities to support trusted collaboration between AI agents. 🔗 theblockbeats.info/flash/343…

3k block sample from 24972500: ⏱️ Max time: 8.18s, average time: 3.69s 📊 Statistics: p50=3.7s, p90=4.3s, p99=5.7s

Our zkDTVM 16-GPU prover is now live and has taken the #1 spot on Ethproofs (@eth_proofs), enabling real-time proving for @ethereum: 🔹99% proving < 5.7s 🔹100% proving < 8.18s 🔹Avg proving time: 3.69s 🔹Security: 100-bit (UDR) Open-sourcing soon. Launching on @JovayNetwork soon. As the trusted on-chain data infrastructure for the agentic era, zkDTVM will power our ecosystem of Agent Web3 solutions—such as agent payments—with a robust, verifiable backbone. openlabs-intl.antdigital.com… #ZK #zkVM #Ethereum #EthProofs #RealTimeProving

Real-world assets, on-chain trust. Our trusted physical asset on-chain solution is deployed in: 🔹EV Charging Infra 🔹Energy Storage 🔹Solar & Renewables 🔹Battery Swapping 🔹Fleet Management We deliver: ✅ End-to-End Trustworthiness ✅ Flexible Deployment ✅ Compliance & Security We will open-source core components to drive industry standardization. 🔗 openlabs-intl.antdigital.com…

Our zkDTVM single GPU prover is live on Ethproofs @eth_proofs! 🔹23s average proving time. 🔹100-bit security level (UDR). Our 16-GPU prover goes live soon. 🔗 ethproofs.org/clusters/12c79…

Glance + speak = payment done. No phone in hand. No Gas to manage. On-chain confirmation in seconds. #Web3 #AI #AgenticPayment #LookToPay #AIGlasses

Introducing "Look-to-Pay": a stablecoin AI glasses payment solution. When Web3 meets AI, you can complete payment with just a glance. Agentic payment. No phone needed. The future of frictionless payments is here.🔗 openlabs-intl.antdigital.com… #Web3 #AI #AgenticPayment #EIP3009 #LooktoPay #SmartGlasses

What if Web3 payments only required a glance? AI glasses scan the WC2 QR code on the POS, triggering on-chain payment. One glance → voice confirmation → transaction completes in seconds. No phone in hand. Low friction. Smooth as Web2, fully on-chain settlement. Powered by USDC on @JovayNetwork. #USDC #Jovay #AIGlasses #Web3 #Stablecoin #AIPayment #SmartGlasses #GaslessPayment #LooktoPay

At the Hong Kong Web3 Festival 2026 @festival_web3, Ant Digital Technologies CTO @frida_ying delivered a keynote speech titled "AI x Web3 Convergence: Building the Trust Layer for the Agentic Economy." As agents become independent economic actors, they face 4 critical cracks: 🔹Prompt Jailbreaks: Logic vulnerabilities & lack of safety boundaries. 🔹Identity Crisis: No KYC/KYB, anonymous wallets, no legal liability. 🔹Payment Deadlock: CAPTCHAs & biometrics block AI from traditional finance. 🔹Trust Deficit: No way to verify strangers in autonomous A2A transactions. We believe Web3 is the trust layer AI needs. @frida_ying introduces our 4R Full-Stack Architecture to rebuild the foundation: 1️⃣ Agentic Runtime (DTClaw): A secure OS for agents. With our CARLI security model, it prevents logic hijacking and ensures financial-grade compliance. It turns risky code into safe, auditable digital labor. 🔗 dtclaw.antdigital.com/home?l… 2️⃣ Payment Rails: Native, programmable settlement. AI can’t pass CAPTCHAs, so we built our Agent Wallet (MPC-based) and gasless micro-payment channels. Money flows like data—fast, autonomous, and compliant. 3️⃣ Agent Registry: On-chain identity via DID (ERC-8004). Every agent gets a unique "Digital Passport" with verifiable reputation and skills. Our Security Guard ensures every A2A handshake is authenticated and safe. 4️⃣ Root Infrastructure (Agentic chain): Powered by Jovay and zkVM. We enable verifiable AI—proving compute correctness without re-running tasks. This solves the scalability and trust issues at the deepest level. The future isn't just AI. It's AI + Trust. We are building the infrastructure where agents are safe, payments are fluid, identities are verified, and compute is trusted. Let's pave the road for the agentic economy together. 🚀

Our Trusted Physical Asset On-Chain Solution ensures data integrity from generation to settlement: 🔹Device Root of Trust 🔹Encrypted Transmission 🔹Oracle Verification 🔹On-Chain Certification A complete trust loop. Check out the use cases & benefits below. 👇 openlabs-intl.antdigital.com…

Today, we are proud to officially launch our Trusted Physical Asset On-Chain Solution. By deeply integrating our Embedded Trusted SDK, IoT Oracle, and the Jovay Stack Layer3, we provide end-to-end trusted infrastructure for physical asset tokenization. Our first overseas IoT asset-to-public-chain project and the world's first trusted IoT asset implementation on a public chain. Our solution tackles two critical challenges: 🔹 Ensuring data authenticity at the source before on-chain submission. 🔹 Efficiently processing massive volumes of device data on-chain. We are about to deploy this in live customer scenarios. 🔗 openlabs-intl.antdigital.com… #IOT #AssetTokenization #DePIN #Layer3 #Jovay

How can we ensure that intelligent systems are safe, trustworthy, and controllable? How can we build a secure and trustworthy transaction and payment system between agents (A2A)? How can agents achieve autonomous evolution and efficient collaboration? We answer these questions in the article below. openlabs-intl.antdigital.com…

Last month at Consensus HK, Ant Digital Technologies CTO @frida_ying delivered a forward-looking keynote on Web3 x AI innovation. Her talk, "Beyond Bolt-On AI: The Rise of AI-Native Web3," outlined Ant Digital Technologies' three-pillar strategy for building the Internet of Agents (IoA) amidst the AI agent explosion: 🚀 Agentic DID: Launching the first identity system tailored for agents. 🚀 Agent Pay: Enabling micropayments & auto-settlements. 🚀 Jovay: An AI-native infra featuring verifiable AI, cross-chain identity, and ultra-high performance. Additionally, Agentar, our financial-grade intelligent agent platform, has launched its co-work/openclaw version, delivering top-tier performance in stability, fund security and full-process controllability. Whatever sector you operate in—finance, energy or DeFi—Agentar is your ideal launchpad. #Agent #agentpayment #agenticDID #Jovay #Agentar #OpenClaw

🚀 Don't hesitate. Come prove your Cairo! Complete CUDA backend and full support for Cairo VM AIRs are coming. Now, 37x faster ZK proving with GPU acceleration. Two months ago, it was only 8x. Check out NitrooZK-stwo-cairo latest update! ⬇ openlabs-intl.antdigital.com…

AntChain Bridge Latest Update! AntChain Bridge, a universal cross-chain framework continuously developed for over three years, supports rapid deployment and ensures secure, reliable, and efficient cross-chain transmission of digital assets and messages. Supported chains: 1. EVM-based chains like #Ethereum, #Base, #Jovay, #Pharos, etc. 2. MoveVM-based chains like #SUI. 3. SVM-based chain like #Solana (coming soon). 4. Other chains like #Dioxide. Check out our progress ⬇ github.com/AntChainOpenLabs/…

Happy new year to you all!