1X Neo
Last reviewed
May 17, 2026
Sources
38 citations
Review status
Source-backed
Revision
v5 · 5,226 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 17, 2026
Sources
38 citations
Review status
Source-backed
Revision
v5 · 5,226 words
Add missing citations, update stale details, or suggest a clearer explanation.
| 1X Neo | |
|---|---|
 | |
| General information | |
| Manufacturer | 1X Technologies |
| Country of origin | Norway / United States |
| Year introduced | 2024 (NEO Beta) |
| Status | Pre-order (shipping 2026) |
| Price | $20,000 (early access) / $499/month (subscription) |
| Height | 167 cm (5 ft 6 in) |
| Weight | 30 kg (66 lb) |
| Degrees of freedom | 75 total (22 per hand) |
| Actuation | Tendon Drive (bio-inspired) |
| Compute | 1X NEO Cortex (NVIDIA Jetson Thor) |
| Battery life | Up to 5.5 hours |
| Noise level | 22 dB |
| Website | 1x.tech/neo |
1X Neo (stylized as NEO) is a bipedal humanoid robot developed by 1X Technologies (formerly Halodi Robotics), a Norwegian-American robotics company. Designed specifically for use in private homes, NEO is a lightweight, human-sized android that combines a bio-inspired tendon-driven actuation system with a soft, pinch-free exterior and near-silent operation. The robot was first introduced as NEO Beta in August 2024, followed by the improved NEO Gamma variant in February 2025. In October 2025, 1X opened consumer pre-orders at a price of $20,000, with initial U.S. deliveries planned for 2026.
NEO represents a deliberate departure from the heavy, industrial-grade humanoids being developed by other companies. Instead of targeting factory floors, 1X has positioned NEO as a consumer product for everyday households, prioritizing safety, quiet operation, and natural human interaction. The robot is backed by significant investment from OpenAI, Samsung Next, EQT Ventures, and Tiger Global, with the company having raised over $125 million across early rounds and reportedly pursuing a $1 billion raise at a $10 billion valuation in late 2025.
In January 2026, 1X unveiled the 1X World Model, a physics-grounded generative video model that lets NEO acquire new tasks by watching video rather than relying solely on explicit programming or teleoperated demonstrations. Three months later, in April 2026, the company opened its NEO Factory in Hayward, California, billed as the most vertically integrated humanoid robot factory in the United States, with a stated capacity of 10,000 units in its first year and an ambition to reach 100,000 units annually by the end of 2027.
1X Technologies was founded in May 2014 in Moss, Norway, by roboticist Bernt Oivind Bornich along with co-founders Phuong Nguyen, Jorgen Sundell, and Pal Loken.[1] The company was originally named Halodi Robotics and focused on developing safe actuators and full-body control systems for industrial and healthcare robotics applications. Bornich, who studied Robotics and Nanoelectronics at the University of Oslo, had been interested in electromechanics since age 11 and envisioned building general-purpose robots that could coexist safely with humans.[2]
The company's early research centered on its proprietary Revo1 motor, a high-torque brushless direct current (BLDC) motor paired with low-gear-ratio cable drives. At the time of its development, the Revo1 was described as the highest torque-to-weight direct drive motor in the world.[3] This motor technology became the foundation for the company's first commercial robot, EVE.
In 2018, 1X (then Halodi Robotics) released EVE, a wheeled, self-balancing humanoid robot designed for logistics, security, and medical environments. Standing 188 cm (6 ft 2 in) tall and weighing 87 kg (192 lb), EVE used wheels rather than legs for locomotion and was capable of traveling at speeds up to 14.5 km/h (9 mph). The robot could carry loads of up to 15 kg and operate for approximately six hours on a single charge.[4]
EVE's most notable commercial deployment came through a partnership with ADT Commercial (Everon), which ordered 140 to 250 EVE units for autonomous night security patrols in commercial buildings across the United States.[5] This deployment represented one of the largest single orders of humanoid robots at the time and provided 1X with crucial real-world operational data. Teleoperated demonstrations performed on EVE later became a primary input to NEO's training datasets, since both robots share many low-level control conventions.
In 2022, Halodi Robotics rebranded as 1X Technologies and shifted its primary focus from industrial and enterprise robotics to domestic consumer robotics. The company began developing humanoid assistants intended for use in private homes, leveraging the data and experience gained from EVE's commercial deployments to train artificial intelligence models for household tasks.[6]
In July 2025, 1X consolidated its teams from Moss, Norway, and Sunnyvale, California, into a new 80,000-square-foot global headquarters in Palo Alto, California, designed to accommodate up to 400 employees. The company maintains manufacturing operations at the Hayward, California facility and in Moss, Norway, with an additional U.S. plant in San Carlos slated to come online later in 2026.[7]
The first version of the NEO platform, called NEO Beta, was unveiled on August 30, 2024.[8] The NEO Beta was a proof-of-concept humanoid standing 165 cm (5 ft 5 in) tall and weighing 30 kg (66 lb). It featured 1X's proprietary Tendon Drive actuation system and hands with 22 degrees of freedom each. The robot could walk at up to 4 km/h and sprint at 12 km/h, with a battery life of two to four hours.
NEO Beta demonstrated the core design philosophy behind the platform: a lightweight, bio-inspired humanoid built specifically for safe operation around people in domestic settings. The robot used tendon-driven actuators rather than traditional rigid gearboxes, creating movements described as inherently gentler and more compliant than those produced by conventional servo motors.[9]
On February 21, 2025, 1X introduced NEO Gamma, a substantially improved iteration of the NEO platform.[10] The Gamma version incorporated several major upgrades over the Beta:
| Feature | NEO Beta | NEO Gamma |
|---|---|---|
| Hardware reliability | Baseline | 10x improvement |
| Operating noise | ~32 dB | ~22 dB (10 dB reduction) |
| Exterior covering | Exposed frame | Knitted nylon suit (machine-washable) |
| Joint safety | Partially covered | Fully enclosed, pinch-proof |
| Gait quality | Basic bipedal walking | Natural human gait with arm swings |
| Movement capabilities | Walking, basic tasks | Squatting, sitting in chairs, running |
| Audio system | Basic speakers | 4-mic beamforming array + 3-speaker system |
| AI model | Early Redwood prototype | Improved Redwood with whole-body control |
The NEO Gamma's exterior is wrapped in a knitted nylon body suit manufactured using a Japanese Shima Seiki machine through a whole-garment seamless knitting process. This 3D-printed nylon fabric conforms to the robot's body without impeding movement and is designed to reduce potential injuries from robot-to-human contact. The suit is machine-washable and available in three colors: tan, gray, and dark brown.[11]
NEO Gamma also introduced "Emotive Ear Rings," a design element intended to improve communication between the robot and its users. The robot's dynamic control skills, including natural walking with arm swings and the ability to squat and sit, are learned through reinforcement learning from human motion capture data, running at 100 Hz.[12]
On October 28, 2025, 1X officially launched NEO as a consumer product and opened pre-orders through its online store.[13] The company described NEO as "the world's first consumer-ready humanoid robot designed to transform life at home." Two purchasing options were offered:
Additionally, customers could place a fully refundable $200 deposit to reserve a unit. 1X announced plans to begin U.S. deliveries in 2026 and expand to international markets starting in 2027.[14]
On April 30, 2026, 1X opened the NEO Factory in Hayward, California, describing it as the first vertically integrated humanoid robot factory in the United States.[33] The facility manufactures the majority of NEO's critical components in-house, including the Revo1-derived motors, batteries, structural skeleton, Tendon Drive transmission assemblies, knitted soft goods, and sensors. According to the company, the Hayward plant can deliver up to 10,000 NEO units in its first operating year, and a second U.S. facility in San Carlos is scheduled to come online later in 2026.
1X has publicly targeted a production rate of more than 100,000 NEOs per year by the end of 2027, with planned increases in automation along the way. The Hayward opening was accompanied by an announcement of 200 or more high-skill American manufacturing jobs and was framed by 1X as a shift from research and development to volume product manufacturing.[34]
NEO stands 167 cm (5 ft 6 in) tall and weighs 30 kg (66 lb), making it one of the lightest bipedal humanoids in development. Its human-proportioned frame is designed to navigate standard home environments, including doorways, staircases, and furniture layouts. The robot's 75 total degrees of freedom enable a wide range of movements across the full body.[15]
| Specification | Value |
|---|---|
| Height | 167 cm (5 ft 6 in) |
| Weight | 30 kg (66 lb) |
| Total degrees of freedom | 75 |
| Degrees of freedom per hand | 22 |
| Walking speed | Up to 4 km/h (2.5 mph) |
| Sprinting speed | Up to 12 km/h (7.5 mph) |
| Payload capacity (lift) | 68 kg (150 lb) |
| Payload capacity (carry) | 25 kg (55 lb) |
| Operating noise | 22 dB |
| Battery capacity | 5,200 mAh |
| Battery life | Up to 5.5 hours (standard mode) |
| Hand IP rating | IP68 (fully submersible) |
| Body IP rating | IP44 (splash-proof) |
| Colors available | Tan, Gray, Dark Brown |
NEO's most distinctive engineering feature is its patented Tendon Drive actuation system, which mimics the human musculoskeletal system. Rather than using traditional direct motor-to-joint coupling with rigid gearboxes, the system uses artificial polymer tendons to transmit force from motors to joints, similar to how biological tendons connect muscles to bones.[16]
This design offers several advantages for a home environment robot:
The Tendon Drive also relocates many motors away from joint axes and toward the torso, redistributing mass closer to the robot's center of gravity and reducing the angular inertia of swinging limbs.
Safety has been a central design priority for NEO, given its intended deployment in homes alongside children, elderly individuals, and pets:
NEO runs on the 1X NEO Cortex, a custom compute module based on the NVIDIA Jetson Thor platform. This module delivers up to 2,070 FP4 TFLOPS of AI compute, providing sufficient processing power to run large language models and vision-language-action models entirely on-device.[17]
| Component | Details |
|---|---|
| Compute module | 1X NEO Cortex (NVIDIA Jetson Thor) |
| AI compute | Up to 2,070 FP4 TFLOPS |
| Cameras | Dual 8.85 MP stereo fisheye cameras (90 Hz) |
| Microphones | 4-microphone beamforming array with echo cancellation |
| Speakers | 3-speaker system (1 chest, 2 pelvis) |
| Connectivity | Wi-Fi, Bluetooth, 5G |
| AI model | Redwood (160M parameter VLA, runs at ~5 Hz on-device) |
The dual stereo fisheye cameras operate at 90 Hz and provide 3D perception for simultaneous localization and mapping (SLAM), enabling NEO to navigate rooms, avoid obstacles, and build spatial maps of its environment. The four-microphone beamforming array supports voice interaction, while the three-speaker system provides AI voice responses and room-filling audio.[18]
NEO's intelligence is powered by Redwood, 1X's proprietary vision-language-action (VLA) model. Redwood is a 160-million-parameter transformer that handles perception, navigation, and manipulation through a single unified architecture rather than treating these as separate subsystems.[19]
Redwood operates as an end-to-end model, processing visual input from NEO's cameras and language instructions from users, then generating motor commands for the robot's entire body. The model runs at approximately 5 Hz on NEO's onboard GPU, meaning it makes five decisions per second about how to move the robot.[20]
Internally, Redwood fuses three modality streams: pre-trained language embeddings from natural language instructions, vision tokens from a pre-trained vision transformer over the fisheye camera feeds, and proprioception embeddings derived from joint positions and applied forces. These streams pass through a stack of transformer blocks that produces a latent representation. The representation is then decoded into continuous action trajectories using a diffusion policy, a generative decoder that samples motor command sequences from a learned conditional distribution rather than predicting a single deterministic output.[35]
A key technical distinction of Redwood is its whole-body control approach. Many robotic systems treat locomotion (walking, balancing) and manipulation (grasping, carrying) as separate problems handled by different controllers. Redwood predicts arm and hand commands together with walking, manipulation, and pelvis pose commands in one shot, enabling more fluid and natural behavior. For example, NEO can bend at the hips and spine to pick up clothes from the floor, or brace one hand against a surface for stability while opening a heavy door with the other.
Redwood is trained on a combination of real-world teleoperation data and autonomous rollouts from both EVE and NEO fleets. Human teleoperators wear VR headsets to remotely control NEO robots, performing household tasks while the system records camera feeds and motor commands. Critically, Redwood learns from both successful and failed task attempts, allowing the model to improve from any interaction regardless of outcome. The failure trajectories are particularly valuable because they teach the policy which actions to avoid in similar contexts.[21]
The model is also trained to handle variation in tasks, such as picking up objects it has never encountered before in locations it has not previously visited. This generalization capability is essential for a home robot, where environments and objects vary enormously from one household to the next. Researchers at 1X have reported emergent multi-contact manipulation, where the robot stabilizes a target with one hand while acting with the other.
The entire Redwood model runs locally on NEO's embedded GPU without requiring a cloud connection. This design choice addresses both latency concerns (the robot must react in real time) and privacy considerations (home environments contain sensitive data). NEO can operate in areas with unreliable internet connectivity, such as basements or gardens, without losing autonomous capability.[22]
On January 12, 2026, 1X publicly unveiled the 1X World Model, an AI system designed to let NEO acquire new tasks by predicting the visual consequences of its own actions rather than relying exclusively on teleoperated demonstrations or pre-programmed behaviors.[36] The release marked a shift in 1X's stated AI roadmap from imitation learning toward a self-supervised paradigm informed by world-model research in reinforcement learning.
The 1X World Model is, at its core, a physics-grounded video generative model. Given current camera frames and a prompt describing a desired outcome, the model produces a short video that depicts a plausible sequence of future observations consistent with the laws of physics in the robot's environment. NEO then runs an inverse dynamics model that maps the generated future frames back into the motor commands needed to actually reach those states.
In practice, this allows the robot to plan in image space: if the prompt is "put the mug on the shelf," the world model imagines what the next several seconds would look like if NEO performed that action, and the inverse dynamics model translates that imagined trajectory into joint targets. Redwood then executes the resulting commands while reacting to discrepancies between the imagined and observed frames.[37]
A key claim by 1X is that the World Model lets NEO learn from internet-scale video footage in addition to teleoperated robot data. Because the model only needs to predict pixels rather than precise robot actions during pre-training, it can be trained on large quantities of unlabeled human video to absorb general priors about how objects, surfaces, and people behave. The robot-specific inverse dynamics model then bridges those imagined dynamics and NEO's body. 1X framed this as moving humanoids closer to the scaling regime that has driven large language models, where most of the training signal comes from passively observed data.
The World Model does not replace Redwood. Instead, 1X positions the two systems as complementary: Redwood provides the reactive, low-latency controller that runs at roughly 5 Hz on-device, while the World Model serves as a planning and self-supervision layer that generates training trajectories and proposes plans for unfamiliar tasks. Real-world failures feed back into the World Model, refining its predictions over time.
In marketing and product documentation around the consumer launch, 1X grouped NEO's behaviors into three modes that customers can invoke through the companion app or by voice:
| Mode | Purpose | Typical autonomy |
|---|---|---|
| Chores Mode | Scheduled or on-demand household tasks such as tidying, laundry, and carrying groceries | Mixed: autonomy plus teleoperated assistance when stuck |
| Companion Mode | Conversation, education, daily reminders, and social interaction | Mostly autonomous, on-device language and dialogue |
| Autonomous Mode | Ambient operation: navigating, monitoring the home, opening doors, operating switches | Fully autonomous within designated areas |
At launch, 1X has publicly characterized NEO's overall autonomy as roughly 60 to 70 percent of attempted tasks completed without operator intervention, with internal targets of 80 to 90 percent by 2027 and more than 95 percent by 2028 as the Redwood and World Model stacks mature.[38]
NEO is designed to perform a range of domestic tasks, including:
During demonstrations, NEO has shown the ability to vacuum, water plants, fold clothes, and navigate rooms without bumping into people or furniture.[23]
As of mid-2026, NEO's autonomous capabilities remain a work in progress. In independent testing by journalists, many household tasks required human assistance through 1X's teleoperation system rather than being completed fully autonomously.[24] The company has been transparent about this limitation, describing NEO's initial capability as "basic autonomy" that will grow over time as the AI models improve through real-world data collection.
To bridge the gap between current AI capabilities and fully autonomous operation, 1X employs a "human-in-the-loop" approach. When NEO encounters a task it cannot handle autonomously, a trained 1X operator wearing a Meta Quest 3 headset can remotely access the robot's cameras and control its movements to complete the task. These teleoperation sessions also serve as training data for Redwood, gradually expanding the range of tasks NEO can perform independently.[25]
1X offers scheduled sessions where an "Expert" teleoperator can assist with complex tasks, and users have privacy controls that determine when and whether a teleoperator can access their robot's cameras and sensors. The company has stated that one human supervisor oversees roughly eight teleoperators at a time, with all sessions logged for incident review.
Because NEO operates inside private homes and can in some cases stream camera and microphone data to remote operators, 1X has built a layered set of privacy controls into the product:
| Control | Purpose |
|---|---|
| No-go zones | Rooms or areas that are off-limits to both autonomy and teleoperation |
| Visual blur | Automatic blurring of human faces and bodies in the operator's video feed |
| Audio masking | Distortion of ambient speech so operators cannot transcribe conversations |
| Time windows | Configurable schedules that restrict when teleoperation is permitted |
| Data sharing opt-out | Per-task consent to contribute recordings back to 1X for training |
| Operator vetting | Background checks and confidentiality agreements for every teleoperator |
| Session logs | Recordings of all teleoperation sessions for incident tracing |
Founder Bernt Bornich has publicly argued that supervised teleoperation is more auditable than an in-person cleaner, since every minute of remote access is logged. Critics counter that even with these controls, the existence of a live camera feed inside the home is a fundamentally new privacy attack surface.[32]
1X Technologies has raised over $125 million in venture capital funding through multiple rounds:
| Round | Date | Amount | Lead investor | Key participants |
|---|---|---|---|---|
| Series A2 | March 2023 | $23.5 million | OpenAI Startup Fund | Tiger Global, Sandwater, Alliance Ventures, Skagerak Capital |
| Series B | January 2024 | $100 million | EQT Ventures | Samsung Next, OpenAI Startup Fund, Tiger Global, Nistad Group |
The Series A2 round in March 2023 was significant as one of the first major investments by the OpenAI Startup Fund into a robotics company, signaling OpenAI's interest in embodied AI and physical intelligence.[26] The $100 million Series B round in January 2024 brought the company's total raised to approximately $137 million and was led by EQT Ventures, a European venture capital firm.[27]
In September 2025, reports indicated that 1X was in discussions to raise up to $1 billion in new funding at a valuation of at least $10 billion, representing a more than 12-fold increase from its approximately $820 million valuation at the time of the Series B round.[28] This potential fundraise, if completed, would make 1X one of the most highly valued humanoid robotics companies in the world.
In December 2025, 1X announced a strategic partnership with EQT to make up to 10,000 NEO humanoid robots available to EQT's global portfolio of more than 300 companies between 2026 and 2030. The partnership focuses on high-impact use cases where robots work closely with humans, including logistics, facility operations, warehousing, manufacturing, and healthcare.[29]
1X plans to launch pilot deployments in the United States in 2026, followed by scaling across Europe and Asia. If fully realized, this deployment would represent one of the largest commercial rollouts of humanoid robots in the industry's history. While NEO was originally designed as a home robot, this partnership demonstrates the platform's versatility in industrial settings.
In March 2025, 1X CEO Bernt Bornich announced plans to begin in-home testing of NEO Gamma units in "a few hundred to a few thousand" homes by the end of 2025.[30] The pilot program relies on a combination of autonomous AI operation and teleoperation to handle tasks, with the primary goal of collecting real-world household data to improve NEO's AI models.
Early adopters in the pilot program effectively serve as data contributors, helping 1X build a diverse dataset of robot-home interactions that is used to train future versions of the Redwood AI model and the 1X World Model. Each hour of household operation, whether autonomous or teleoperated, produces training signal that flows back into model updates pushed over the air to deployed units.
NEO occupies a unique position in the humanoid robotics market as one of the few robots designed primarily for consumer home use rather than industrial or enterprise applications.
| Robot | Manufacturer | Height | Weight | Target market | Price | Status (2026) |
|---|---|---|---|---|---|---|
| NEO | 1X Technologies | 167 cm | 30 kg | Home / consumer | $20,000 | Pre-order, shipping 2026 |
| Optimus | Tesla | 173 cm | 57 kg | Industrial / consumer | ~$20,000-$30,000 (projected) | In development |
| Figure 03 | Figure AI | 170 cm | 70 kg | Industrial / enterprise | Not announced | In development |
| Atlas | Boston Dynamics | 150 cm | 89 kg | Research / industrial | Not for sale | Research platform |
| GR-2 | Fourier Intelligence | 175 cm | 63 kg | Healthcare / industrial | ~$90,000+ | Limited availability |
NEO differentiates itself through several factors: its exceptionally low weight (30 kg compared to 57 kg or more for competitors), its bio-inspired tendon actuation for safe human interaction, and its explicit consumer home focus. Most competing humanoids are significantly heavier and designed for industrial or research applications. As of 2026, NEO is the only consumer-focused humanoid actively shipping to homes.[31]
However, the company faces stiff competition from well-funded rivals. Tesla has projected a similar $20,000 to $30,000 price range for its Optimus humanoid, though a consumer launch date has not been confirmed. Figure AI has raised billions in funding and is developing advanced humanoids with large language model integration. 1X is the first of these rivals to operate a vertically integrated U.S. consumer humanoid factory and ship paid units to private addresses.
NEO has received attention for its ambitious consumer home positioning and its lightweight, safety-first design approach. The partnership with OpenAI and the high-profile funding rounds have generated significant media coverage. The January 2026 unveiling of the 1X World Model and the April 2026 opening of the Hayward factory further amplified press attention, with several outlets describing 1X as the most consumer-ready humanoid company in the West.
Critics and analysts have raised several concerns. The reliance on teleoperation for many tasks means that early adopters are essentially paying $20,000 for a robot that cannot yet perform most household tasks independently. Privacy advocates have noted that the teleoperation model requires human operators to view live camera feeds from inside people's homes, even with user-controllable access settings.[32] The long-term viability of the $499/month subscription model has also been questioned, given the robot's current limitations. Some commentators have characterized the launch as "selling the dream," arguing that the visible autonomy gap could damage consumer trust if not closed quickly.
Supporters counter that the human-in-the-loop approach is a pragmatic strategy for collecting the training data needed to build truly autonomous home robots, and that early adopters are investing in a platform that will improve substantially over time through software and AI updates. They also point to the company's published autonomy roadmap, which projects steady improvements through 2028, as evidence that the teleoperation phase is intended to be transitional rather than permanent.