您正在使用IE低版浏览器，为了您的雷峰网账号安全和更好的产品体验，强烈建议使用更快更安全的浏览器
此为临时链接，仅用于文章预览，将在时失效
0
导语：截至目前，蚂蚁灵波开源模型在GitHub累计斩获超3.3万颗star，稳居国内具身模型前列。
10月27日，由蚂蚁灵波科技主办的首届“蚂蚁灵波具身大模型挑战赛”线上初赛正式落下帷幕。官方数据显示：大赛共吸引了海内外1855支队伍、3252位选手报名。参赛者来自13个国家，覆盖648 个知名高校、科研院所以及头部科技企业及具身智能公司，其中包含 41个海外高校和单位，首届大赛就创下行业最大规模。据悉，线下总决赛将于11月14至16日在上海张江以48小时真机黑客松的形式举行。
1855支团队 清北及海外名校与产业一线选手同台竞技
今年9月，蚂蚁灵波科技联合魔搭社区、阿里云天池，在2026外滩大会上发起了首届“蚂蚁灵波具身大模型挑战赛”。这是一场具身智能算法挑战赛，以开源模型LingBot-VLA2.0为技术底座，聚焦模型训练调优、仿真泛化测评与真实机器人适配，面向全球企业、高校、个人开发者及技术爱好者开放。
10月27日0点，线上初赛报名截止。大赛共有1855支团队报名，吸引了清华大学、北京大学、上海交通大学、浙江大学、中国科学技术大学、同济大学、中国科学院大学、哈尔滨工业大学，以及香港大学、香港科技大学、香港中文大学等知名高校的学子参赛，卡内基梅隆大学、佐治亚理工学院、南加州大学、伦敦大学学院、慕尼黑工业大学、新加坡国立大学、南洋理工大学、悉尼大学、瑞典皇家理工学院、早稻田大学等国际名校研究力量也参与其中。
产业从业者也将本次大赛作为技术“练兵场”。华为、腾讯、阿里、字节跳动、美团、微软、比亚迪、小米等一线科技公司开发者纷纷组队应战；优必选、傅利叶智能、智元机器人、逐迹动力、北京人形机器人创新中心等国内人形机器人与具身智能企业也悉数下场打擂，让本届挑战赛成为兼具学术前沿探索与产业工程落地的实战场。
大赛组委会公布的画像显示：在全部参赛者中，高校及研究人员占比超七成（74.7%），其中硕博研究生比例高达 67.4%；同时，来自企业和一线实战领域的开发者占比达到 25.2%，高学历、强实战成为本届选手的显著标签。
半数选手手握VLA实战经验 加速“具身大脑”走向物理世界
从参赛选手的技术图谱来看，这批开发者高度集中于具身智能的最核心技术链路。其中，VLA（视觉-语言-动作）或多模态大模型背景选手占比高达 51.7%；深度学习训练占比 48%；强化学习（31.9%）、仿真环境（25.4%）、模型部署（23.4%）、真机调试（20.9%）、机器人控制（20.1%）紧随其后。
这些数据表明，蚂蚁灵波具身大模型挑战赛不仅吸引了算法研究人才，更沉淀了大量具备“上机、真动、真调”实操能力的复合型工程开发者。
业内分析人士指出，蚂蚁灵波首届具身大模型挑战赛能在短时间内吸引超1800支全球高水平队伍，一方面印证了具身智能赛道正迎来前所未有的技术爆发期，开发者求知若渴；另一方面，也展现出开发者社区对蚂蚁灵波自研具身大模型的高度认可。
截至目前，蚂蚁灵波开源模型在GitHub累计斩获超3.3万颗star，稳居国内具身模型前列。海外主流技术社区也已实现对灵波模型的适配。LingBot-VLA、LingBot-VA 均已完成对全球最大机器人开发平台 LeRobot 的生态适配，LingBot‑Vision 合入业界主流视觉骨干模型库 timm 主线。LingBot-World 成为全球首个同时获得 vLLM、SGLang 两大主流开源推理引擎支持的世界模型。
通过开放前沿的基模能力、真实的物理仿真与工程环境，蚂蚁灵波不仅为全球开发者提供了一个竞技舞台，更加速了具身智能“通用大脑”从理论算法走向全场景产业赋能的进程。
按照赛程，更具挑战的线下决赛暨蚂蚁灵波首届开发者大会将于11月14日至16日在上海举办，15支队伍以48小时真机黑客松的形式进行角逐。届时，各位选手将在真实世界碰撞出怎样的创新火花，值得期待。
为了您的账户安全，请验证邮箱
请验证您的邮箱
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
You are using a low-end IE browser and strongly recommend a faster and safer browser for your Thunder Peak account security and better product experience.
This is a temporary link, only for article preview, which will expire at the time
0
Introduction: To date, the ants peppo-source model has accumulated more than 33,000 stars in GitHub and is at the forefront of the domestic model.
On October 27, the opening of the first "Ant-Link-Eat-Eat-Eat LLM Challenge" competition, hosted by Ant-Link Technology, officially ended. Official data show that the competition attracted a total of 1855 teams and 3,252 candidates, both at home and abroad. The contestants came from 13 countries and covered 648 leading institutions of higher learning, scientific research institutes and head-of-the-art technology enterprises and smart companies, including 41 overseas institutions and units, with the largest number of industries in the first competition. It is known that the online finals will be held in Zhang Jiang, Shanghai, from 14 to 16 November, in the form of 48 hours of real-time hacking.
1855 teams, leading schools and industry first-line contestants in Qingbei and overseas.
In September of this year, the antspod technology community, the Alibaba Cloud pool, launched the first LLM Challenge at the 2026 Outer Beach Congress. This is a smart algorithm challenge, with the open source model LingBot-VLA2.0 as the technology base. Focus model training is fine-tuned, simulation broad-based and real robotic, open to businesses, universities, personal developers and technology lovers worldwide.
On 27 October, at 0:00, the registration for the online primary competition was closed. A total of 1,855 teams were enrolled in the tournament, which attracted the participation of the students of the Qinghua University, Beijing University, Shanghai Transport University, Zhejiang University, China University of Science and Technology, Toji University, Chinese Academy of Sciences University, Harbin University of Technology, Hong Kong University of Science and Technology, Chinese University of Hong Kong, and the participation of the research forces of leading international schools such as Carnegie Mellon University, Georgia Polytechnic Institute, University of Southern California, London University College, Munich Industrial University, Singapore National University, South Asia University of Technology, Sydney University, Royal College of Technology of Sweden, and the University of Harada.
Industry practitioners have also used this competition as a technical “drainground”. China, Tencent, Ali, Z113Z, the United States of America, Microsoft, Biadi, and Mi, the first-line technology company developers, have been in teams; the human robots and smart companies, such as Yoo-Bing, Fuliers, Ji-won robots, Modified Dynamics, Beijing Human Robotic Innovation Center, and smart companies, have all come to battle, making this challenge a real battleground for exploration and industrial engineering as well.
The picture published by the organizing committee shows that the proportion of all contestants is over 70 per cent (74.7 per cent), of whom 67.4 per cent are graduate students, while the share of developers from business and the front-line field is 25.2 per cent, with a high level of education and strong combat becoming a prominent label for the contestant.
Half the players hold VLA field experience and accelerate the brain to the physical world.
Based on the technical profiles of the contestants, the developers are highly concentrated on the most intelligent and core technology links. Of these, VLA (visual-linguistic-action) or multi-modular LLM background players accounted for 51.7 per cent; in-depth learning training for 48 per cent; intensive learning (31.9 per cent); simulation environment (25.4 per cent); model deployment (23.4 per cent); real machine debugging (20.9 per cent); and robotic control (20.1 per cent).
These data suggest that the antspoon LLM Challenge has attracted not only algorithms, but also a large number of complex engineering developers with “onboard, real-time, real-temporal” capability.
Industry analysts have noted that the first ants LLM Challenge will attract more than 1800 high-level teams globally in a short period of time, attesting, on the one hand, to the unprecedented technological boom and the thirst for knowledge on the part of developers, and, on the other, to the high level of recognition of LLM in the development community.
To date, the ants open source model has accumulated more than 33,000 stars in GitHub and is in the forefront of domestic models. The adaptation of the Lingbo model has also been achieved by mainstream technology communities overseas. LingBot-VLA, LingBot-VA have all completed the ecological adaptation of LeRobot, the largest robot development platform worldwide, LingBot-Vision, into the industry's mainstream visual backbone model library timm main line. LingBot-World is the first world model to be supported by the two main mainstream open-source reasoning engines, v LLM and SGLang.
Through an open front line of foundational modelling, real physical simulations and engineering environments, the ant-film wave not only provides a competitive arena for global developers, but also accelerates the movement of the smart “generic brain” from theoretical algorithms to full-situ industry empowerment.
In accordance with the course, the more challenging bottom-line finals and the first antspo development conference will be held in Shanghai from 14 to 16 November, with 15 teams competing in the form of 48 hours of real-time hackers. At that time, it is worth looking forward to what innovative sparks the players will encounter in the real world.
For your account security, please verify the mailbox.
Please verify your mailbox.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.