01.AI has reached a partnership with Hangzhou City and Linping District, with state-owned capital at both city and district levels jointly investing a combined several hundred million yuan. The company's headquarters will be located in Linping District, Hangzhou, which will serve as the foundation for building an industrial AI transformation base. The partnership comes at a critical stage as the large-model company founded by Kai-Fu Lee accelerates commercialization and advances its Pre-IPO strategic positioning; it had previously set targets of 15 hundred million yuan in order value by 2026 and an IPO by 2027.
At the 2026 Zhiyuan Innovation Day, Zhiyuan Co-founder Yao Maoqing unveiled the new-generation embodied foundation model GO-3 Preview. The company says the model has topped the three major evaluation leaderboards—RoboDojo, RoboTwin, and RoboColiseum—and plans to fully open-source it by the end of 2026. The model adopts Zhiyuan's proprietary embodied-native architecture WRAM (World Reasoning Action Model), which unifies the modeling of understanding, prediction, and action, and combines the Genie Evolver post-training system with the Genie Master agent framework to form an end-to-end learning system spanning pre-training, post-training, and post-deployment.
TMTPost, citing Yicai, reports that Meta has partnered with Chinese robotics company Fourier to build a humanoid robot prototype in about three months, with the two parties said to already be developing a second version. Codenamed Max internally, the robot weighs about 48 kg and is used for Meta internal research and experiments. Fourier founder Gu Jie said it was not convenient to comment, and the partnership has yet to be officially confirmed.
At the Embodied Intelligent Robotics Conference of the SAECCE on 28 10 2026, Gu Jie, Vice President of XPeng Group, disclosed that the solid-state battery designed specifically for the IRON humanoid robot has completed small-batch production rollout, with the robot itself in the final stage before formal mass production. The battery reduces weight by approximately 30% and increases capacity by approximately 30% compared with conventional liquid lithium-ion batteries, and features a thermal runaway propagation prevention design. According to XPeng's earlier plan, IRON will enter volume production by the end of 2026 and be launched and delivered in 2027.
Microsoft AI released the image editing model MAI Image 2.7 as a preview on 10 27, 2026. Officially, it ranks No. 2 on the Artificial Analysis Image Editing leaderboard and No. 3 on the Arena Edit leaderboard, and outperforms the 2.6 version across all 10 use case categories and 7 edit operations covered by the evaluation. The update focuses on control in complex editing, covering style adjustments, relighting, text modification, and asset repair.
Xiaomi has unveiled its new-generation bionic dexterous hand, featuring 21 degrees of freedom, 26 drive motors, and 1500 tactile sensing points, with a palm weight of 324 grams. Its design is based on motion analysis of 1827 standard operating procedures across 16 typical workstations in an automotive factory, adopting a hybrid tendon and direct-drive structure with most motors concentrated in the forearm. This marks Xiaomi's latest step in advancing humanoid robot development driven by production needs.
According to People's Weekly, the 5-day 2026 World Robot Conference closed in Beijing Yizhuang on 8 23, with 373 companies exhibiting over 3000 products and 311 new product debuts. UBTECH, Leju, Sinnade, and others showcased industrial humanoid robots in real on-site operations, while Robot Era's logistics solution is now in regular operation at SF Express and China Post logistics centers. Data released alongside the event show that in H2026, China's humanoid robot shipments exceeded 4 units, a global share of 97%.
At the Zhiyuan Innovation Day on 29 10, 2026, Xiong Yan, Partner and Co-President of Zhiyuan (Agibot), launched Lingqu OS1.0, the industry's first embodied-intelligence-native operating system. The system defines a new hardware access paradigm for robot embodiments, supporting rapid access to 100+ types of peripherals across 15 categories. Addressing requirements for high reliability, determinism, and low latency, it completed full-stack deterministic latency optimization, with official figures showing end-to-end latency jitter under 1000Hz high-frequency task loads of less than 4 microseconds. Lingqu OS1.0 plans to be fully open-sourced by the end of 2026.
According to TMTPost, Tencent Hunyuan's simultaneous interpretation large model Hy-SimulMT was released and open-sourced on 10 26. The model emphasizes low latency and high accuracy, supports real-time Chinese-English speech-to-speech translation, and features real-time voice cloning to reproduce the speaker's voice and tone. It debuted on Tencent's Hy Translation APP with a free experience.
Baidu has released and open-sourced Ernie Unlimited OCR 2, which processes ultra-long documents through Sparse Reference Sliding Window Attention (SR-SWA) and the Unlimited-Horizon Decoding Paradigm. According to the technical report, in hundred-page single-sequence tests, peak KV cache dropped from 47GB to 1.77GB, a reduction of about 96%; 16-way concurrent throughput increased from 360 to 1315 tokens per second. A real-world test on the 884-page The Lord of the Rings completed parsing in about 43 minutes, with MPDocBench and MarathonDocBench overall scores of 84.59 and 77.06 respectively, both the highest among evaluated models.
On Beijing time 10 30, SenseTime open-sourced the multimodal model SenseNova Loop-9B-Preview, built on a Looped Transformer architecture: roughly 9B parameters, with 32 layers of weights shared and looped to perform 56 layers of computation per forward pass, officially claiming 27B-level performance along with 8 benchmark comparisons. Weights are on Hugging Face and code on GitHub under Apache 2.0. Community attention focuses on the choice of comparison targets as well as looped inference latency and KV cache overhead; independent testing and the technical report have yet to appear.