# GigaAI

> Source: https://aiwiki.ai/wiki/gigaai
> Updated: 2026-07-07
> Categories: Humanoid Robots, Robotics Companies
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution to "AI Wiki (aiwiki.ai)".

**GigaAI** (also written as **Giga AI**, Chinese: 极佳视界 / 极佳科技, pinyin Jijia Shijie) is a Chinese artificial intelligence company headquartered in [Beijing](/wiki/beijing), [China](/wiki/china), building what it calls "Physical AGI": foundation models, world models, and humanoid hardware that let robots and vehicles act in the real world without task-specific programming. Founded in 2023 by [Huang Guan](/wiki/huang_guan), a Tsinghua University PhD in automation, the company describes itself as China's first startup focused on world model research for physical AI.[1][2] Its stack pairs the [GigaBrain](/wiki/gigabrain) vision-language-action (VLA) foundation model and the [GigaWorld](/wiki/gigaworld) world model platform with two robots: the [Maker H01](/wiki/maker_h01) wheeled humanoid for industry and the SeeLight S1 wheeled humanoid for the home. Backed by [Huawei](/wiki/huawei) (through Huawei Habo Investment) and a wide group of Chinese state and venture funds, GigaAI raised about 3.5 billion yuan (roughly $490 million) across its Pre-B, B1, and B2 rounds inside three months in 2026, reaching a valuation above 10 billion yuan (roughly $1.4 billion).[3][15][16]

## History

Giga AI was founded in 2023 and is based at Number 8, Courtyard 1, Zhongguancun East Road, Haidian District, Beijing, putting it close to the campuses of [Peking University](/wiki/peking_university) and [Tsinghua University](/wiki/tsinghua_university) in the city's main technology corridor.[1]

Founder Huang Guan holds a PhD from Tsinghua University's Department of Automation. Before starting the company he led visual perception research at [Horizon Robotics](/wiki/horizon_robotics), served as a partner at PhiGent Robotics, and earlier worked at Microsoft Research Asia and Samsung Research China.[2] The other named founders round out a strong research bench: co-founder and chief scientist Zhu Zheng earned his PhD from the Chinese Academy of Sciences' Institute of Automation in 2019, did postdoctoral work in Tsinghua's automation department from 2019 to 2021, has published more than 50 papers with over 13,000 citations, and was listed among Stanford's global top 2% of scientists for 2022, 2023, and 2024; co-founder Sun Shaoyan previously served as an Alibaba Cloud director and as general manager of Horizon Robotics' data closed-loop product line.[2] According to Gasgoo, the core team also draws on former vice-president-level architects from large Chinese internet firms, recipients of Huawei's "Genius Youth" recruiting program, and hardware leads from established Chinese robotics companies.[2]

The company positioned itself early on as an "embodied foundation model and general robotics" firm, and publicly described its ambition as becoming "the OpenAI of the physical world": building general models that let a single robot or vehicle handle many tasks, in the same way [ChatGPT](/wiki/chatgpt) handles many text tasks.[2]

GigaAI's first commercial traction came in autonomous driving. In March 2025 the company's DriveDreamer world model was integrated into ECARX's AutoGPT in-vehicle AI platform, a deal announced through ECARX's Nasdaq investor releases.[4] DriveDreamer is positioned as the first world model trained on real driving data, used to generate rare-case driving scenarios in simulation to cut down on physical road testing.[4][5]

In October 2025 GigaAI signed a partnership with the Hubei Humanoid Robot Innovation Center to build what the two sides called the world's first "world-model-driven virtual-real embodied intelligence data factory," a facility that produces synthetic and real training data for embodied AI at scale.[6]

A few weeks later, in early November 2025, GigaAI closed a hundred-million-yuan Series A1 round co-led by [Huawei Habo Investment](/wiki/huawei_habo_investment) and Huakong Fund. The deal marked Huawei's first direct equity investment in a Chinese physical AI startup of this kind and was widely reported as a signal of Huawei's pivot from purely software-defined AI toward world-model-based physical AI.[3][7]

Later the same month, on or around November 12, 2025, GigaAI officially unveiled the Maker H01, its first fully self-developed wheeled humanoid robot, branding it a "Physical AGI Native Platform" and using language about a "ChatGPT moment" for robotics.[8][9]

On February 2, 2026, roughly two months after the launch, GigaAI announced the start of large-scale Maker H01 deliveries. The first unit went to the Hubei Humanoid Robot Innovation Center, with subsequent units shipped to customers in automotive manufacturing, 3C electronics, warehousing and logistics, high-end guiding, and home terminal scenarios. The company set a public target of around 1,000 Maker H01 deliveries during 2026.[6][2]

On March 5, 2026, GigaAI closed a nearly 1 billion yuan (~$140 million) Pre-B financing round. The investor list included Shanghai Pudong Science and Technology Investment, SMIC Juyuan, the Shanghai Semiconductor Industry Investment Fund, Linxin Capital, Xingyuan Capital, Wanlin International, CICC Capital, Su VC, Huaqiang Capital, Changjiang Capital, Optics Valley Industrial Investment, and Xishan State Investment, with several earlier backers (CICC Capital, Huaqiang Capital, Caixin Capital, Zhangke Yaokun) doubling down.[2][3][15]

At the end of March 2026 GigaAI's GigaWorld-1 world model took first place on the WorldArena benchmark, a Tsinghua-led evaluation, becoming the only embodied world model to pass 60 points overall.[20] The company then raised a roughly 1.5 billion yuan Series B1 round in April 2026, unveiled the SeeLight S1 home humanoid robot in May 2026, and closed a roughly 1 billion yuan Series B2 round around June 2026. Across the Pre-B, B1, and B2 rounds GigaAI took in about 3.5 billion yuan (roughly $490 million) in three months and reached a valuation above 10 billion yuan (roughly $1.4 billion), making it one of the best-funded startups in China's embodied AI sector.[16][17]

## Products

GigaAI builds along three connected layers it calls "foundation model, body, and scenario." The foundation model is GigaBrain; the body is the Maker and SeeLight robots; the world model and data factory that feed both is GigaWorld.[6][2]

### Maker H01

The [Maker H01](/wiki/maker_h01) is GigaAI's flagship wheeled humanoid robot and the company's main piece of industrial hardware. It is designed as a native platform for foundation-model control rather than an existing robot retrofitted for AI.

| Specification | Details |
|---------------|---------|
| Type | Wheeled dual-arm humanoid |
| Height | 160 cm |
| Weight | 64 kg |
| Total DOF | 28 (excluding end effectors) |
| Vertical reach | 0 to 2 meters |
| Arm DOF | 7 per arm (bionic) |
| Arm payload | 5 to 10 kg per arm |
| End effectors | Modular: grippers or 5-finger dexterous hands (7 DOF each) |
| Head cameras | 5 RGB + 4 RGB-D (distributed across head, chest, hands) |
| Chassis | Omni-directional, all-wheel-drive |
| LiDAR | 360-degree |
| Max speed | 8 km/h |
| Battery runtime | About 4 hours per charge |
| Ingress protection | IP20 |
| Materials | Aluminum alloy frame with ABS composite shells |
| Software | Integrated full-stack, controlled by GigaBrain |
| Launch date | November 2025 |
| Indicative price | ~$160,000 |

Sources: Humanoid.guide product database and the November 2025 launch coverage.[8][9]

#### Sensor configuration

The Maker H01 carries a deliberately heavy sensor stack so that GigaBrain can fuse multiple viewpoints in real time.

| Location | Sensors |
|----------|---------|
| Head | RGB cameras + RGB-D cameras |
| Chest | RGB cameras + RGB-D cameras |
| Hands | RGB cameras + RGB-D cameras |
| Chassis | 360-degree LiDAR |

The nine cameras (5 RGB plus 4 RGB-D) spread across head, chest, and hands let the robot see what it is manipulating, the obstacles around it, and the people in the room at the same time. The 360-degree LiDAR on the chassis handles longer-range navigation and obstacle avoidance.[8]

#### Modular end effectors

Each arm ends in a standardized attachment point. Customers can swap between simple grippers for pick-and-place tasks and 5-finger dexterous hands with 7 DOF per hand for more delicate manipulation such as laundry folding or pouring. Per-arm payload is rated at 5 to 10 kg depending on the end effector and posture.[9]

#### Target applications

In the company's own framing, the Maker H01 is built for industrial, commercial, and home scenarios. Public reporting on the first wave of deliveries lists automotive manufacturing, 3C electronics, warehousing and logistics, high-end guiding, and home terminal use cases. Independent product databases also mention hospitality reception, workplace assistance, light logistics, and education.[6][9]

### SeeLight S1

The SeeLight S1 is GigaAI's home humanoid robot, unveiled in May 2026 and pitched by the company as China's first general-purpose household humanoid robot. Like the Maker H01 it is a two-armed, wheeled platform, but it is aimed at domestic chores rather than factory work, and its intelligence runs on the GigaWorld world model so that it can look at a cluttered room, infer what needs doing, and plan its own actions rather than follow step-by-step scripts.[18][19]

The SeeLight S1 was developed with the Hubei Humanoid Robot Innovation Center and the Hubei Humanoid Robotics Industry Alliance, the same state-backed partners behind GigaAI's data factory.[19]

#### What can the SeeLight S1 do?

In public demonstrations the SeeLight S1 has been shown chopping vegetables, frying eggs, loading a washing machine, hanging laundry, making a bed, and opening curtains.[18][19] For safety in the home the robot uses what GigaAI describes as a "compliant control" system that immediately stops its motion when it makes contact with a person or a pet, and the company says the first customers it is targeting are households with elderly members, children, or pets.[18]

#### When will the SeeLight S1 ship, and how much will it cost?

GigaAI plans to deploy the first 100 SeeLight S1 units into its own employees' homes at the end of May 2026, then hand units free of charge to families in Wuhan, the capital of Hubei province, in the first half of 2027, before a retail launch around June 2027.[18][19] The company is targeting a hardware price below 100,000 yuan (about $14,700) at that launch, roughly half the robot's cost at unveiling.[18] Zhu Zheng, a GigaAI co-founder, told Wuhan's Changjiang Daily that he expects household robots to make "significant breakthroughs in both commercialisation and embodied AI model capabilities by 2028."[18] A next-generation model, the SeeLight S2, is planned as a follow-up.[16]

### GigaBrain

[GigaBrain](/wiki/gigabrain) is GigaAI's vision-language-action (VLA) foundation model for robots. The publicly released version, GigaBrain-0, takes RGB-D input (color plus depth) and outputs both low-level actions and an "Embodied Chain-of-Thought" intermediate representation that the model uses to reason about object positions, contact points, and task order before acting.[10][11]

A few features stand out:

- The model is trained largely on world-model-generated synthetic data, including video generation, real-to-real transfer, human-to-robot transfer, view transfer, and sim-to-real transfer, which the GigaBrain-0 paper says lets training use up to 90% synthetic data without losing real-world performance.[10][11]
- A technique the authors call Knowledge Insulation keeps the chain-of-thought head from interfering with the low-level action head during training.
- In the paper's benchmarks, GigaBrain-0 outperforms Physical Intelligence's [Pi0](/wiki/pi0) (π0) on dexterous tasks like laundry folding and paper towel preparation, long-horizon tasks like juice preparation and table bussing, and mobile manipulation tasks like moving boxes and laundry baskets.[10][11]

A lightweight variant, **GigaBrain-0-Small**, is optimized for edge inference on NVIDIA Jetson AGX Orin and is intended for deployment on robots in the field rather than in the cloud.[11]

#### What is GigaBrain-0.5M?

GigaBrain-0.5M is a later, world-model-trained version of the VLA that the company calls the first embodied model to achieve efficient learning and "self-evolution" through [reinforcement learning](/wiki/reinforcement_learning).[15] It builds on GigaBrain-0.5, which was pretrained on more than 10,000 hours of real-world robot data, and adds a training loop the authors name RAMP (Reinforcement leArning via world Model-conditioned Policy): instead of reacting only to what it currently sees, the policy is conditioned on futures predicted by a video world model, so it can plan longer-horizon tasks.[21] In the paper, an intermediate version (GigaBrain-0.1) ranks first on the RoboChallenge benchmark with a 51.67% average success rate, and RAMP delivers roughly a 30% improvement over the RECAP baseline on hard tasks like box packing and laundry folding.[21] GigaAI also hosts the GigaBrain Challenge at CVPR 2026, which uses its data and benchmarks to compare external VLA submissions.[12]

### GigaWorld

GigaWorld is GigaAI's world model platform, used as a data engine for embodied AI. It generates synthetic driving and robot interaction data through video generation and transfer learning techniques. The GigaWorld-0 paper describes it as "world models as a data engine," and the public results argue that VLA models trained largely on GigaWorld-generated data can hit strong real-world performance without much physical robot interaction during training.[13]

The Hubei "virtual-real embodied intelligence data factory" runs on this stack, mixing synthetic data from GigaWorld with real teleoperation and demonstration data collected from Maker robots on factory floors.[6]

#### How does GigaWorld rank against Google and NVIDIA?

At the end of March 2026 GigaAI reported that its GigaWorld-1 model had taken first place on WorldArena, a world-model benchmark built by Tsinghua University together with several other leading institutions that scores systems on 16 core indicators plus three real-world application tasks.[20] GigaWorld-1 was described as the only embodied world model to break through 60 points on the benchmark's overall score, with a physics-adherence result about 16% ahead of the second-place entry and 3D accuracy close to a perfect score, ahead of entries from Google and NVIDIA.[20] GigaAI framed GigaWorld-1 as an "action-conditioned" world model that couples explicit action modeling with a differentiable physics engine, and said its related GigaWorld-Policy delivered roughly ten times the inference speed and training efficiency of mainstream alternatives while lifting task success rates by about 30%.[20]

### DriveDreamer

DriveDreamer and its 4D successor DriveDreamer-2/4D are GigaAI's autonomous driving world models. They generate diverse driving scenarios, including edge cases, in simulation. ECARX integrated DriveDreamer into its AutoGPT in-vehicle AI application in March 2025, with the stated goal of reducing reliance on physical road testing and shortening development cycles for new ADAS features.[4][5] [Li Auto](/wiki/li_auto) is named in Chinese reporting as another deployment customer.[2]

## Technology approach

GigaAI's technical pitch sits on a few connected ideas:

### World models as a data engine

Rather than collecting enormous amounts of real robot interaction data, GigaAI argues that high-quality world models can generate that data in simulation. GigaWorld renders synthetic robot trajectories and driving scenes that GigaBrain then trains on. The company claims this is how it gets around the standard bottleneck in embodied AI, namely that real-robot data is slow and expensive to collect.[10][13] Founder Huang Guan has argued that VLA-dominated embodied models suffer from two flaws, inefficient model architectures and inefficient real-world data collection, and that "the rapid development of the world model provides an unprecedented solution" to both.[15]

### Physical AGI native platform

The Maker H01 is built from the ground up to be controlled by a foundation model. Sensor placement, the bionic 7-DOF arm geometry, the omni-directional chassis, and the compute architecture are all chosen to feed and be fed by GigaBrain rather than retrofitted onto a robot designed for hand-written control software.[8][9]

### Mass production and supply chain

GigaAI puts a lot of public emphasis on its ability to actually build robots at volume. Coverage of the February 2026 delivery announcement notes that the company stresses its "systemic capabilities in engineering mass production, supply chain integration, and fast-paced software-hardware iteration," the kinds of capabilities that distinguish a shipping company from a research lab.[6]

### Connected to autonomous driving

Unusually for a humanoid robotics company, GigaAI also sells into autonomous driving. The DriveDreamer family is in production use at ECARX (and reportedly Li Auto), and the same world model approach is meant to transfer between cars and robots. Huawei's interest, per the Pandaily and Gizmochina reporting, is partly about applying world models to both autonomous driving and embodied robotics inside Huawei's wider ecosystem.[3][7]

## Funding

GigaAI has raised multiple rounds from Chinese state-backed funds, private venture capital, and Huawei's investment arm. Between the November 2025 Series A1 and mid-2026 the company disclosed four named rounds, with cumulative investment well above 4 billion yuan and a valuation above 10 billion yuan (roughly $1.4 billion) by the Series B2.[16][17]

| Round | Approximate amount | Date | Notable investors |
|-------|--------------------|------|-------------------|
| Series A1 | Hundred-million-yuan level | Early November 2025 | Huawei Habo Investment, Huakong Fund (co-leads) |
| Pre-B | ~1 billion yuan (~$140 million) | March 5, 2026 | Shanghai Pudong Science and Technology Investment, SMIC Juyuan, Shanghai Semiconductor Industry Investment Fund, Linxin Capital, Xingyuan Capital, Wanlin International, CICC Capital, Su VC, Huaqiang Capital, Changjiang Capital, Optics Valley Industrial Investment, Xishan State Investment |
| Series B1 | ~1.5 billion yuan (~$210 million) | April 2026 | Multiple Chinese state and venture funds |
| Series B2 | ~1 billion yuan (~$140 million) | Around June 2026 | A Singapore-based fund (Lion City Capital), the China-Belgium Direct Equity Investment Fund, auto-parts maker Wanxiang Qianchao, and others |

Several earlier investors (CICC Capital, Huaqiang Capital, Caixin Capital, Zhangke Yaokun) re-upped in the Pre-B, which is generally read as a positive signal about internal performance.[2][3] Together the Pre-B, B1, and B2 rounds brought in about 3.5 billion yuan (roughly $490 million) inside three months, and GigaAI has said the Series B2 proceeds go toward its data and algorithm systems, its physical AGI core models, and expansion into both household and industrial robot markets.[16]

## Partnerships and customers

GigaAI's announced partners cover both robotics and autonomous driving:

| Partner | Domain | Nature of relationship |
|---------|--------|------------------------|
| [Huawei](/wiki/huawei) (via Huawei Habo) | Investment, strategic | Series A1 co-lead; exploring cooperation across Huawei business lines |
| [ECARX](/wiki/ecarx) | Autonomous driving | DriveDreamer integrated into AutoGPT in-vehicle AI (March 2025) |
| [Li Auto](/wiki/li_auto) | Autonomous driving | DriveDreamer deployment (per Chinese reporting) |
| Hubei Humanoid Robot Innovation Center | Robotics | Joint "virtual-real embodied intelligence data factory" (October 2025); first Maker H01 delivery customer (February 2026); co-developer of the SeeLight S1 |
| FAW Tooling Die Technology | Automotive manufacturing | Maker H01 deployment partnership |
| Longsheng Technology | Robotics / manufacturing | Agreement to deploy about 1,000 Maker H01 units over three years in Wuxi |

The first wave of Maker H01 commercial deployments is concentrated in automotive plants, 3C electronics factories, logistics warehouses, and a small number of high-end guiding and home use deployments, and GigaAI says it works with more than 30 automakers and autonomous driving firms across its robot and DriveDreamer businesses.[6][2][17]

## Reception and context

GigaAI is part of a broader 2025-2026 wave of Chinese humanoid robotics startups, alongside [Unitree](/wiki/unitree), [AgiBot](/wiki/agibot), [Galbot](/wiki/galbot), and [UBTECH](/wiki/ubtech). Gasgoo's tracking of the sector reports that embodied intelligence companies are now routinely raising rounds at the 1-billion-yuan-plus level, with the highest single rounds in the segment topping 3 billion yuan; after its rapid Pre-B, B1, and B2 sequence in 2026 GigaAI sits among the best-capitalized names in the group, at a valuation above 10 billion yuan.[14][16][17]

Pandaily and Gizmochina both frame the Huawei investment as the more strategically interesting moment, since Huawei rarely takes direct equity in this kind of physical AI company. They argue the bet is on a shift away from purely language-model-driven AI and toward systems that map vision and physical signals directly to actions in cars and robots.[3][7]

Independent product databases caution that the Maker H01 is still in early production. Humanoid.guide rates the robot at a manipulation performance score of 4/10, lists its glass-to-action latency as 250 to 450 ms, and notes that as of the listing the unit is best treated as an early-production platform aimed at integrators rather than a consumer-ready home robot.[9]

## See also

- [Maker H01](/wiki/maker_h01)
- [GigaBrain](/wiki/gigabrain)
- [GigaWorld](/wiki/gigaworld)
- [DriveDreamer](/wiki/drivedreamer)
- [Embodied AI](/wiki/embodied_ai)
- [Humanoid robot](/wiki/humanoid_robot)
- [World model](/wiki/world_model)
- [Vision-Language-Action Model](/wiki/vision-language-action_model)
- [Huawei](/wiki/huawei)
- [Unitree](/wiki/unitree)
- [AgiBot](/wiki/agibot)

## References

1. Tracxn. "Giga AI: 2026 Company Profile, Team, Funding & Competitors." tracxn.com.
2. Gasgoo Auto News. "With 1 billion yuan invested, Gigaai aims to become the 'OpenAI of the physical world.'" March 2026. https://autonews.gasgoo.com/articles/market-industry/with-1-billion-yuan-invested-gigaai-aims-to-become-the-openai-of-the-physical-world-2030133640946626561
3. Pandaily. "Huawei Habo Invests in Physical AI Startup 'GigaAI,' Expanding into the World Model Frontier." November 2025. https://pandaily.com/huawei-habo-invests-in-physical-ai-startup-giga-ai-expanding-into-the-world-model-frontier
4. ECARX Holdings / GlobeNewswire. "ECARX Integrates DriveDreamer World Model into AutoGPT." March 5, 2025. https://www.globenewswire.com/news-release/2025/03/05/3037263/0/en/ECARX-Integrates-DriveDreamer-World-Model-into-AutoGPT.html
5. Nasdaq. "ECARX Holdings Inc. Enhances AutoGPT AI Application with GigaAI's DriveDreamer Integration to Accelerate Autonomous Driving Development." March 2025.
6. Gasgoo Auto News. "GigaAI Begins Large-Scale Deliveries." February 2026. https://autonews.gasgoo.com/articles/news/gigaai-begins-large-scale-deliveries-2018322126438551553
7. Gizmochina. "Huawei's Surprise Investment Could Redefine the Future of Robots and Self-Driving Cars." November 14, 2025. https://www.gizmochina.com/2025/11/14/huaweis-surprise-investment-could-redefine-the-future-of-robots-and-self-driving-cars/
8. RoboHub (X). "GigaAI officially unveils its new wheeled humanoid: the Maker H01." November 2025. https://x.com/XRoboHub/status/1993604526449074473
9. Humanoid.guide. "Maker H01 by GigaAI: Wheeled Humanoid Robot." https://humanoid.guide/product/maker-h01/
10. Wang et al. "GigaBrain-0: A World Model-Powered Vision-Language-Action Model." arXiv:2510.19430, October 2025. https://arxiv.org/abs/2510.19430
11. open-gigaai. "giga-brain-0" GitHub repository. https://github.com/open-gigaai/giga-brain-0
12. GigaAI Research. "GigaBrain Challenge 2026 @ CVPR 2026." https://gigaai-research.github.io/GigaBrain-Challenge-2026/
13. "GigaWorld-0: World Models as Data Engine to Empower Embodied AI." arXiv:2511.19861. https://arxiv.org/html/2511.19861v1
14. Gasgoo Auto News. "10 Rounds at 1-Billion Yuan Level, Highest Over 3 Billion: Embodied Intelligence Enters an Era of 'Billion-Yuan Valuation.'" 2026. https://autonews.gasgoo.com/articles/news/10-rounds-at-1-billion-yuan-level-highest-over-3-billion-embodied-intelligence-enters-an-era-of-billion-yuan-valuation-2052998789747601408
15. 36Kr Global. "Jijia Shijie Secures 1 Billion Yuan in Pre-Series B Financing, Driven by 'World Model' to Build World's Leading 'Embodied Schema.'" March 2026. https://eu.36kr.com/en/p/3709674264883587
16. DigitalPhablet. "GigaAI Raises $518M in 3 Months for Chinese Embodied AI & Robotics." 2026. https://digitalphablet.com/news/gigaai-raises-518m-in-3-months-for-chinese-embodied-ai-robotics/
17. Dermot McGrath. "The China Round." Zengen Labs Research, June 19, 2026. https://research.zengenlabs.com/p/the-china-round-june-19-2026
18. South China Morning Post. "Commercial humanoid robots in China may soon do laundry, make beds, care for elderly." May 21, 2026. https://www.scmp.com/tech/article/3354371/commercial-humanoid-robots-china-may-soon-do-laundry-make-beds-care-elders
19. Interesting Engineering. "Chinese firm eyes 2027 home robot launch with Wuhan pilot and sub-$15k pricing." 2026. https://interestingengineering.com/ai-robotics/chinese-firm-eyes-2027-home-robot-launch-with-wuhan-pilot-and-sub-15k-pricing
20. National Center for Science and Technology Innovation (ncsti.gov.cn). "GigaAI's embodied world model tops the global WorldArena leaderboard." April 1, 2026. https://www.ncsti.gov.cn/kjdt/scyq/zgckxc/zgcdt/202604/t20260401_242656.html
21. GigaAI. "GigaBrain-0.5M*: a VLA That Learns From World Model-Based Reinforcement Learning." arXiv:2602.12099, February 2026. https://arxiv.org/html/2602.12099