Spirit AI

Spirit AI (officially Spirit AI (Hangzhou) Technology Co., Ltd.; Chinese: 千寻智能; Qianxun Intelligence) is a Chinese robotics company headquartered in Hangzhou, Zhejiang, China, specializing in embodied intelligence robots and vision-language-action models. Founded in January 2024 by Han Fengtao, AI scientist Gao Yang and Zheng Lingyin (now chief operating officer),^[21][26] Spirit AI develops general-purpose humanoid robots and the foundation models that control them. The company's flagship products include the Moz1 wheeled humanoid robot, the Moz2 commercial service robot shown publicly in July 2026, and the Spirit family of embodied intelligence foundation models.^[23][24] Spirit v1.5 took the top ranking on the RoboChallenge real-world robotics benchmark in January 2026, surpassing the pi0.5 model developed by US-based Physical Intelligence; its successor, Spirit v1.6, topped the RoboArena leaderboard in June 2026, becoming the first Chinese model to lead that NVIDIA, Stanford, and UC Berkeley backed benchmark and outscoring NVIDIA's own Cosmos 3 policy.^[17][22]

In the first half of 2026 alone, Spirit AI raised more than 4.5 billion yuan across four financing rounds, which the venture-capital outlet PEDaily called a record fundraising pace for China's embodied-AI sector; PEDaily reported that the February 2026 rounds took its valuation past 10 billion yuan and the April 2026 round past 20 billion yuan.^[21] In September 2026 Reuters reported that the roughly 300-person company had raised more than 670 million US dollars since its founding and was valued at 20 billion yuan (about 2.9 billion US dollars).^[23] Backers include Shunwei Capital (Lei Jun's fund), Yunfeng Capital (Jack Ma's fund), JD.com, Sequoia China, Prosperity7 Ventures, CATL, Huawei, Xiaomi, TCL Capital, and 360 Capital.

History

Spirit AI was founded in early 2024 (registration documents and most English-language sources cite February 2024; several Chinese profiles list January 2024) by Han Fengtao, who previously co-founded the collaborative-robot maker Rokae Robotics (also transliterated as LuoShi or Luoshi Robotics) in 2015 and served as its chief technology officer.^[1][2] At Rokae, Han led the development of roughly 100 industrial robot models and shipped more than 20,000 units, giving Spirit AI an unusually mass-production-literate founding team for an embodied-AI startup.^[2]

Han was joined by Gao Yang, a UC Berkeley PhD who trained under Pieter Abbeel and is an assistant professor at Tsinghua University's Institute for Interdisciplinary Information Sciences, where he directs the Tsinghua Vision and Embodied Intelligence Laboratory. Gao is the company's chief scientist and led earlier academic work on the ViLa and CoPa models for vision-language grounding in manipulation. A third co-founder, Zheng Lingyin, serves as chief operating officer.^[2][3]

The company's first robot, the Moz0, appeared in July 2024 as an internal development platform. In March 2025, Spirit AI publicly demonstrated its Spirit v1 vision-language-action model by showing a robot completing a full clothes-folding sequence, then widely described as the first Chinese demonstration of long-horizon flexible-object manipulation by an end-to-end VLA system.^[3] The commercial-grade Moz1 humanoid robot followed in June 2025.^[4]

In the second half of 2025 the company expanded rapidly, growing its data-collection workforce. Its relationship with JD.com began in July 2025, when JD led the roughly 600 million yuan Pre-A+ round; the two companies formalized a strategic partnership on March 19, 2026, covering 2026 through 2029, under which Moz robots were deployed at JD MALL stores to demonstrate teleoperated barista work using JD's JoyAI and JoyInside technology. The demonstrations doubled as a data-collection loop, capturing teleoperation trajectories and force feedback to fine-tune the company's models.^[5][6][15]

On December 17, 2025, Spirit AI's robot, branded "Xiaomo" in Chinese and "Moz" in English, was deployed on what CATL described as the world's first humanoid-robot power-battery production line at the company's Luoyang base in Henan province.^[7][8] Five weeks later, in January 2026, Spirit v1.5 topped the RoboChallenge real-world robotics benchmark with a score of 66.09, surpassing the pi0.5 model from Physical Intelligence; Spirit AI simultaneously open-sourced the model and supporting resources.^[9][10] Between late February and early April 2026, Spirit AI announced a series of funding rounds totaling close to 3 billion yuan within about 30 days; PEDaily reported that the February rounds pushed its valuation past 10 billion yuan and the April round past 20 billion yuan, making it one of the highest-valued private embodied-AI companies in China.^[11][12][21]

In May 2026, Spirit AI signed a strategic partnership with Bosch China to industrialize a general-purpose robot "Universal Brain," pairing its vision-language-action models with Bosch's industrial ecosystem.^[19][20] On June 3, 2026, the company announced a 1.5 billion yuan A+ round alongside news that its new Spirit v1.6 model had topped the RoboArena benchmark, outscoring NVIDIA's freshly launched Cosmos 3. The A+ round capped four financing rounds in three months totaling more than 4.5 billion yuan, which Chinese venture-capital outlets described as a record pace for the country's embodied-AI sector.^[19][21]

Around the same time the company added two supply-chain partners. On May 20, 2026 it announced a strategic partnership with the edge-computing chip company D-Robotics, which had integrated Spirit v1.5 with its Sunrise S600 embodied-AI computing platform (rated by D-Robotics at 560 TOPS INT8),^[27] and on June 3, 2026 it signed a strategic cooperation agreement with the motion-technology company Schaeffler covering embodied-AI data collection, core humanoid-robot components (including a long-term procurement relationship), and industrial applications.^[26] At the 2026 World Artificial Intelligence Conference in Shanghai, which opened on July 17, 2026, Spirit AI gave the first public showing of Moz2, a commercial service robot, and demonstrated Moz1 completing a multi-step "tidy up the living room" task from a single instruction using Spirit v1.6.^[24] In August 2026 its Moz1 robots won one event and placed second in another at the second World Humanoid Robot Games.^[28] In a September 2026 Reuters interview, co-founder Gao Yang predicted a GPT-3-style breakthrough for robot "brains" as soon as mid-2027.^[23]

Products

Moz1

The Moz1 is Spirit AI's first commercial-grade humanoid robot, unveiled in June 2025. It is a wheeled humanoid rather than a bipedal one, and it is positioned for industrial, logistics, and service-sector tasks rather than as a household consumer product.^[23] Reuters reported in September 2026 that Spirit AI had tens of Moz1 robots deployed on production lines at CATL and JD.com.^[23]

The Moz1's joints combine motor, gearbox, encoder, and torque sensor into a single module the company designs in-house. Han Fengtao has argued that a robotics company that does not build its own VLA model cannot tell what "good hardware" looks like; the actuator specification reflects that view, with joints tuned for the contact-rich tasks the Spirit foundation models try to learn.^[2][4]

Data collection

Moz1 ships with a wearable data-capture system that Spirit AI uses to collect demonstrations across factories, homes, and labs. The fifth-generation rig, in use by early 2026, captured what the company described as 95 percent usable data at roughly one-tenth the cost of conventional teleoperation. As of February 2026 Spirit AI had accumulated more than 200,000 hours of real-world interaction data from internet video, teleoperation, and wearables, with a stated 2026 target above 1 million hours and a collection team scaling to about 1,000 people.^[11][13]

Later descriptions of the collection operation vary by source and date. At the June 2026 BAAI Conference the company said it had set up more than 300,000 data-collection points across China, staffed by more than 1,000 full-time data-collection specialists.^[25] In a July 24, 2026 company post, Spirit AI described more than 1,000 wearable collection devices online at the same time across more than 100 cities, with more than 1,500 collection staff.^[29] Reuters reported in September 2026 that the company employed around 1,000 contractors nationwide who used wearable data-collection equipment in households and on production lines, and that a Reuters witness at the company's Beijing data training centre saw dozens of young people fitted with sensors repeating motions such as opening fridges, unlocking safes and cutting vegetables with knives.^[23] Reuters noted that operators at other Chinese training facilities may need to repeat a movement more than 50 times to get one "clean" movement; Gao said Spirit AI had found that more diverse "dirty data" let its models improve faster.^[23]

Moz2

Moz2 is a commercial service robot that Spirit AI first showed publicly at WAIC 2026 in July 2026, marking the company's move into retail, hospitality, and office-building settings. The company describes it as its first exploratory product for commercial services, designed together with Spirit AI's embodied-intelligence system and its uDAS wearable data-collection device so that human operation data and Moz1's skills can be transferred to it.^[24]

The company frames Moz1 and Moz2 as sharing one technology stack under a stated roadmap of "industry first, commercial services next, home last."^[24]

Spirit v1 and Spirit v1.5 foundation models

Spirit v1, released in March 2025, was a vision-language-action model demonstrated on long-horizon manipulation tasks such as folding garments. Spirit v1.5, open-sourced in January 2026, is the successor and the model that took the top RoboChallenge ranking.

In explaining the architecture, Han Fengtao said Spirit v1.5 "integrates perception, reasoning and action into an end-to-end system, reducing errors associated with established modular approaches."^[9][10] The team has publicly noted a scaling-law pattern: roughly a tenfold increase in training data yields a step-function improvement of about nine points on internal benchmarks, which is the rationale for the aggressive million-hour data target.^[6][13]

How does Spirit v1.6 compare to NVIDIA Cosmos and pi0.5?

Spirit v1.6, announced on June 3, 2026, is the successor to Spirit v1.5 and the model that made Spirit AI the first Chinese team to top RoboArena, a real-world robot-policy leaderboard co-developed by NVIDIA, Stanford University, and UC Berkeley that scores how effectively a generalist policy handles object manipulation, navigation, tool use, perception, planning, and adaptation to unfamiliar environments.^[17][22] Spirit v1.6 posted a RoboArena score of 1,924, ahead of NVIDIA's Cosmos3-Nano-Policy at 1,881 and NVIDIA's DreamZero at 1,763. The result landed just two days after NVIDIA unveiled its Cosmos 3 world-model family at Computex in Taipei, and was widely covered in the English-language technology press as the first time a Chinese model had led the benchmark.^[17][18][22]

The company positioned the back-to-back RoboChallenge and RoboArena wins as validation of its data-centric thesis, arguing that large volumes of diverse real-world interaction data, rather than raw model scale, drive real-world robot performance.^[17][21]

Spirit AI describes Spirit v1.6 as combining a vision-language-action model with a world model in one architecture that links perception, task understanding, action planning, and prediction of future world states.^[24][25] At the June 2026 BAAI Conference, Gao Yang presented this approach as "The Omni Model," which treats VLA and world-model training as different objective functions within the same model, and reported an internal finding that each tenfold increase in pretraining data for the Spirit models cut the post-training fine-tuning data needed for downstream tasks by a factor of 2.5.^[25] Demonstrations of Spirit v1.6 on Moz1 at the conference and at WAIC 2026 included desk organizing, a capsule-toy machine task, pill sorting, and a four-subtask living-room tidying sequence in which the robot added newly introduced objects to its plan.^[24][25]

Technology

Spirit AI's stack is full-stack by design, mixing foundation-model research with motor and joint engineering. Core areas:

• Vision-language-action models. The Spirit v1 and v1.5 models accept RGB camera input and natural-language task descriptions and emit joint-level motor commands. The team has compared this end-to-end approach to LLM training pipelines, with pretraining on unlabeled video, supervised fine-tuning on teleoperation, and reinforcement learning to clean up failure modes.^[2][6]
• Integrated force-controlled joints. Each Moz1 joint is a self-designed module incorporating motor, harmonic gearbox, dual encoders, and a torque sensor. The company claims industry-leading power density.^[3][4]
• Whole-body control (WBC). A high-frequency WBC layer mediates between the VLA model's commanded behaviors and the joint-level controllers, handling balance and constraint enforcement.^[4]
• Zero-latency teleoperation. A custom rig lets a human operator drive Moz1 in real time, capturing demonstrations that feed the data pipeline.^[4][13]
• Dirty-data pipeline. Co-founder Gao Yang has framed the data approach as "dirty data," arguing that diverse, lightly curated interaction data from wearables, video, and teleoperation generalizes better than smaller curated lab datasets.^[11][13]
• Hourly model iteration. The combination of wearable capture, in-house training, and an end-to-end model lets the team push new checkpoints into testing on roughly an hourly cadence.^[3][4]

Earlier academic work by Gao Yang, including the ViLa visual-language model and the CoPa component-constraint model, feeds into Spirit AI's perception stack.^[3]

Research published by Gao Yang's team at Spirit AI in 2026 includes Legato, a training-time continuation method for action-chunked, flow-matching VLA policies (arXiv, February 2026; accepted to RSS 2026 according to the company), which the authors report beats Real-Time Chunking across five real-world manipulation tasks with about 10 percent improvements in both trajectory smoothness and task completion time.^[30] A July 2026 paper studied co-training with legacy data on a wheeled humanoid platform across two hardware generations with different cameras and grippers; it found that old data gave no measurable benefit until the upgraded robot reached a minimum level of task competence, after which gains rose sharply (flower-insertion success went from 23.3 percent to 86.7 percent with co-training) before tapering near saturation.^[31]

Deployments and demonstrations

CATL battery production line

The largest production deployment is at CATL's Zhongzhou base in Luoyang, Henan, announced on December 17, 2025 and described by CATL as the world's first humanoid-robot power-battery production line.^[7][8][14] Moz robots handle End-of-Line (EOL) and Direct Current Resistance (DCR) tests, which involve plugging high-voltage test connectors into specific points on a battery pack. The job is hazardous when performed by humans because the test plugs carry hundreds of volts.

Reported performance from CATL and Spirit AI:

The deployment ran on CATL-supplied batteries and used the Spirit VLA model to adjust grip and pose in real time when connector positions varied between battery packs.^[7][8][14]

JD MALL retail demonstrations

Under the strategic partnership with JD.com signed in March 2026, Moz robots were placed in JD MALL physical stores, where they performed barista tasks including cup pickup, bean grinding, espresso extraction, and latte service. The robots used JD's JoyAI large model and JoyInside teleoperation platform, and every interaction fed multimodal sensor data, joint trajectories, and force feedback back into Spirit AI's training pipeline. Han Fengtao has framed the deployment as much as a data flywheel as a product launch.^[5][6] The two companies also said they would explore robots for automated sorting and dispensing at JD Pharmacy.^[5]

Bosch China partnership

In May 2026, Spirit AI signed a strategic partnership with Bosch China to jointly industrialize a general-purpose robot "Universal Brain," pairing Spirit AI's vision-language-action models with Bosch's industrial ecosystem.^[19][20] The agreement covers a closed real-world data loop inside Bosch's China factories and logistics centers, integration of Bosch sensors and actuators to speed validation and mass production, and benchmarking of Spirit's models against industrial standards. Han Fengtao said the alliance "marks our transition from technical excellence to industry leadership," adding that "Bosch's resources will scale our Universal Brain architecture." Liu Min, a Bosch China vice president, said Spirit AI had "demonstrated leading technical prowess in embodied AI models" and that the collaboration would "establish a new ecosystem paradigm for robotics."^[19]

Competitions and benchmark results

Spirit AI's robots participated in the 2025 Intelligent Robot Competition in China, completing tasks such as desktop tidying, waste disposal, and microwave operation. In January 2026, Spirit v1.5 topped the RoboChallenge leaderboard, evaluated across 30 real-world manipulation tasks including object placement, target recognition, and tool use; it was the only model to clear the 50 percent task-success-rate threshold at release.^[2][9][10] Five months later, in June 2026, the successor Spirit v1.6 topped the RoboArena leaderboard with a score of 1,924, ahead of two NVIDIA entries, making Spirit AI the first Chinese team to lead that benchmark.^[17][22] At the second World Humanoid Robot Games in August 2026, Spirit AI entered Moz1 robots running Spirit v1.6 in fully autonomous mode, without teleoperation after an initial task instruction. According to the company, they won the office "comprehensive service" scenario event (setting up a meeting room, printing and delivering documents, and shredding files) as the only fully autonomous team, and took second place in a catering event that required taking orders, heating food, and making drinks within 20 minutes.^[28]

Funding

Spirit AI raised at least seven rounds of equity capital between August 2024 and June 2026. The first half of 2026 alone brought four rounds totaling more than 4.5 billion yuan, which the venture-capital outlet PEDaily (投资界) called a record fundraising pace for China's embodied-AI sector, and PEDaily reported that the February rounds lifted the valuation above 10 billion yuan (about 1.4 billion US dollars at the time) and that the April round lifted it above 20 billion yuan.^[11][21] Reuters reported in September 2026 that the company had raised more than 670 million US dollars since its 2024 founding and was valued at 20 billion yuan (about 2.9 billion US dollars); Gao Yang declined to comment on any plans for an initial public offering.^[23] Round nomenclature has been reported inconsistently across outlets: the June 2026 raise was widely described as an "A+" round, which implies the February and April 2026 raises together constituted the company's Series A.

The February 2026 round was reported in English-language press as roughly 290 million US dollars and characterized by PR Newswire and other outlets as a fundraise to scale Spirit AI's "dirty data" pipeline.^[11][13][15] The April 2026 round, co-led by Lei Jun's Shunwei Capital and Jack Ma's Yunfeng Capital, was widely reported as the first time the two funds jointly led an embodied-AI investment.^[12][13][21] The June 3, 2026 A+ round of 1.5 billion yuan was announced the same day Spirit AI reported its Spirit v1.6 RoboArena win; the raise drew first-tier US-dollar funds, large industrial investors, and state-owned funds, with proceeds earmarked for next-generation foundation-model research, a global real-world data-infrastructure buildout, and multi-industry commercialization.^[21]

Strategic investors with industry rather than pure-financial motivations include CATL (battery production deployments), JD.com (retail and logistics), Huawei and Xiaomi (consumer-electronics and silicon ecosystem), TCL (appliance manufacturing), and Minghui Investment, the family office of Inovance Technology chairman Zhu Xingming (industrial automation).^[21] Founder Han Fengtao reportedly held about 24.86 percent of the company directly and 16.39 percent indirectly as of mid-2025.^[2][3]

Leadership

Core team members are drawn from UC Berkeley, Carnegie Mellon University, Tsinghua University, Peking University, Xiaomi, Huawei, and Tencent.^[3]

Strategy and outlook

Spirit AI splits its resources heavily in favor of model and data work, with Han saying the company puts roughly 80 percent of resources into the data system and model rather than hardware program-management.^[6][13] The stated commercial path is to enter narrow vertical markets first (battery assembly, retail service, laundry-style flexible manipulation) and use them as data flywheels for a more general model, with broader deployment of service robots expected within two to three years of Spirit v1.5.^[9][10] The May 2026 Bosch alliance and the June 2026 A+ round both frame the same goal: converting benchmark leadership into industrial deployments and a self-reinforcing real-world data pipeline.^[19][21]

Han has compared the state of embodied AI in 2026 to LLMs in 2023: a moment when raw funding and benchmark performance become threshold requirements rather than differentiators.^[2] The 2026 commercial targets reported by executives included 100 to 150 million yuan in revenue and roughly 200 to several-hundred Moz robot shipments, with larger volumes contingent on customer-side production-line redesigns.^[2][16]

At the June 2026 BAAI Conference, Han argued that the gap between model capability and deployment cost meant conditions for large-scale commercialization were not yet mature, so the industry's priority for the year should be large-scale pretraining and accumulating high-quality data; he said "data is the real bottleneck facing the industry today."^[25]

In an interview with Reuters at the company's Beijing offices on September 17, 2026, Gao Yang said that "the brain is indeed the weakest link in the complete robotics stack" and that an industry breakthrough comparable to OpenAI's GPT-3 could come around mid-2027. He said Spirit AI's robots had reached a 90 percent success rate on simple tasks in structured living-room environments and set out a staged timeline: "The next one to two years mark the initial window for industrial applications. Two years from now, we'll see robots deployed in commercial service settings doing simpler tasks. Entering homes is far harder than both." Reuters reported the company's view that home deployment could take at least eight years because model development faces a data bottleneck. Gao said fine-motor actions such as unscrewing a bottle cap, and tasks the robot has not seen, remain difficult, and explained the company's reliance on real-world data over simulation: "Simulators handle rigid bodies well, but flexible objects like deformable electric cables remain a problem."^[23]

Disambiguation

This article covers the Chinese robotics company Spirit AI / Qianxun Intelligence. It is unrelated to Spirit AI Ltd., a UK-based interactive-storytelling and dialogue-tools company known for the Character Engine middleware used in some games. The two companies share an English brand name but have no corporate, technical, or personnel overlap.

See also

• Moz1
• Embodied AI
• Vision-language-action model
• Humanoid robot
• CATL
• Bosch
• Physical Intelligence
• pi0.5
• RoboChallenge
• RoboArena
• World Humanoid Robot Games
• NVIDIA
• Rokae Robotics
• Tsinghua University
• UC Berkeley
• Pieter Abbeel
• JD.com
• Shunwei Capital
• Yunfeng Capital

References

1. ^Baidu Baike (English). "Spirit AI (Hangzhou) Technology Co., Ltd." baike.baidu.com/en/item/Spirit%20AI%20(Hangzhou)%20Technology%20Co.,%20Ltd./33601.
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^936Kr. "Qianxun Intelligence's Valuation Exceeds 10 Billion Yuan in Just Two Years, Yet It Dares Not Relax." eu.36kr.com/en/p/3701216103281408. 2026.
3. ^1 ^2 ^3 ^4 ^5 ^6 ^736Kr. "Four Industry Leaders' Capital Joins Forces to Create a Tens-Billion-Worth Unicorn in Embodied AI Field." eu.36kr.com/en/p/3698397830164098. 2026.
4. ^1 ^2 ^3 ^4 ^5 ^636Kr. "Qianxun Intelligence Completes 528 Million Yuan in Pre-A Round Financing to Accelerate the Implementation of Embodied AI General Models." eu.36kr.com/en/p/3224908715539591. 2025.
5. ^1 ^2 ^3Gasgoo. "Spirit AI, JD Group Form Strategic Partnership." autonews.gasgoo.com/articles/news/spirit-ai-jd-group-form-strategic-partnership-2034655625756172289. March 19, 2026.
6. ^1 ^2 ^3 ^4 ^536Kr. "Meet Generalist at the Peak: How Did Qianxun Intelligence Secure $3 Billion in Just 30 Days?" eu.36kr.com/en/p/3756066027209477. 2026.
7. ^1 ^2 ^3Spirit AI. "A World First! Embodied AI Robots Achieve Large-Scale Deployment in CATL Battery Production Line." spirit-ai.com/en/news/23. December 2025.
8. ^1 ^2 ^3South China Morning Post. "China's CATL marks trailblazing deployment of humanoid robots at scale on the factory floor." scmp.com. December 18, 2025.
9. ^1 ^2 ^3 ^4Xinhua News Agency. "Chinese AI startup tops global embodied intelligence benchmark." english.news.cn. January 13, 2026.
10. ^1 ^2 ^3 ^4People's Daily Online. "Chinese AI startup tops global embodied intelligence benchmark." en.people.cn/n3/2026/0114/c90000-20413808.html. January 14, 2026.
11. ^1 ^2 ^3 ^4 ^5AsiaTechDaily. "Spirit AI Raises $280M to Scale 'Dirty Data' Robotics, Is China Betting Big on Embodied AI?" asiatechdaily.com. February 2026.
12. ^1 ^2Wedoany. "Embodied AI Tech Company Qianxun Intelligence Completes Two Funding Rounds Totaling 3 Billion RMB Within 30 Days." en.wedoany.com/shortnews/135429.html. April 2026.
13. ^1 ^2 ^3 ^4 ^5 ^6 ^7PR Newswire. "Spirit AI Lands $280M to Scale Embodied AI Through 'Dirty Data'." prnewswire.com. February 25, 2026.
14. ^1 ^2Caixin Global. "CATL Tests Humanoid Robots on Battery Production Lines." caixinglobal.com. December 19, 2025.
15. ^1 ^2AIBase. "Sentient Intelligence Company Qianxun Intelligence Completes Pre-A+ Round Financing of Nearly 600 Million Yuan." aibase.com/news/19806. July 2025.
16. ^KrAsia. "This startup thinks large models will finally make robots useful." kr-asia.com. 2025.
17. ^1 ^2 ^3 ^4 ^5The Next Web. "Spirit AI beats Nvidia on RoboArena robotics benchmark." thenextweb.com/news/spirit-ai-beats-nvidia-roboarena-physical-ai. June 5, 2026.
18. ^eWeek. "Spirit AI, a Chinese Robotics Startup, Tops Nvidia in Physical AI Ranking." eweek.com/news/chinese-startup-spirit-ai-nvidia-physical-ai-ranking-apac. June 2026.
19. ^1 ^2 ^3 ^4 ^5PR Newswire. "Spirit AI and Bosch Partner on General-Purpose Robot 'Universal Brain'." prnewswire.com/news-releases/spirit-ai-and-bosch-partner-on-general-purpose-robot-universal-brain-302765614.html. May 2026.
20. ^1 ^2Gasgoo. "Spirit AI and Bosch China Sign Strategic Partnership to Jointly Promote Large-Scale Deployment of Embodied Intelligence." autonews.gasgoo.com. May 2026.
21. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10PEDaily (投资界). "3 Months, Nearly 5 Billion: Qianxun Breaks the Embodied-AI Fundraising Record." news.pedaily.cn/202606/564786.shtml. June 3, 2026.
22. ^1 ^2 ^3 ^4South China Morning Post. "A Chinese robotics start-up beat Nvidia on a global AI ranking. Is a new tech war brewing?" scmp.com/tech/article/3355838. June 2026.
23. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Laurie Chen, Reuters. "Chinese 'robot brain' startup sees ChatGPT-style breakthrough as soon as next year" (first published as "Founder of Chinese startup Spirit AI says robot brains set for 2027 breakthrough"). reuters.com/world/asia-pacific/founder-chinese-startup-spirit-ai-says-robot-brains-set-2027-breakthrough-2026-09-18/. September 18, 2026.
24. ^1 ^2 ^3 ^4 ^5 ^6Spirit AI. "Hardcore Long-Horizon Tasks × Adorably Approachable! Spirit AI Steals the Spotlight at WAIC as Moz2 Wins Over the Crowd on Its Public Debut." spirit-ai.com/en/news/48. July 2026.
25. ^1 ^2 ^3 ^4 ^5Spirit AI. "2026 Beijing Academy of Artificial Intelligence Conference | Beyond the Debate Over Technical Routes, Spirit AI Bets on Real-World Model Evolution." spirit-ai.com/en/news/43. June 2026.
26. ^1 ^2Spirit AI. "Spirit AI and Schaeffler Announce Strategic Partnership to Advance the Embodied Intelligence Industry." spirit-ai.com/en/news/46. June 3, 2026.
27. ^Spirit AI. "Spirit AI and D-Robotics Announce Strategic Partnership, Bringing Spirit v1.5 to the Sunrise S600 for Deployment Across Diverse Real-World Scenarios." spirit-ai.com/en/news/45. May 20, 2026.
28. ^1 ^2Spirit AI (千寻智能). "全自主推理夺冠！千寻智能斩获世界人形机器人运动会一金一银" (Fully autonomous reasoning wins: Spirit AI takes a gold and a silver at the World Humanoid Robot Games). spirit-ai.com/news/49. August 2026.
29. ^Spirit AI (千寻智能). "北京市委书记尹力，市委副书记、市长殷勇调研千寻智能" (Beijing leaders visit Spirit AI). spirit-ai.com/news/40. July 24, 2026.
30. ^Liu, Yufeng, et al. "Learning Native Continuation for Action Chunking Flow Policies." arXiv:2602.12978. February 2026. arxiv.org/abs/2602.12978.
31. ^Wang, Tao, et al. "When Does Legacy Data Start to Help? Emergent Transfer in Cross-Configuration Robot Learning." arXiv:2607.25593. July 2026. arxiv.org/abs/2607.25593.