1X Technologies NEO Gamma
Last reviewed
May 17, 2026
Sources
23 citations
Review status
Source-backed
Revision
v3 · 5,185 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
May 17, 2026
Sources
23 citations
Review status
Source-backed
Revision
v3 · 5,185 words
Add missing citations, update stale details, or suggest a clearer explanation.
1X NEO Gamma is a humanoid robot developed by 1X Technologies, a Norwegian-American robotics company. Unveiled on February 21, 2025, NEO Gamma is the consumer-ready evolution of the NEO platform, succeeding the NEO Beta prototype that debuted in August 2024. Standing 1.65 meters tall and weighing approximately 30 kilograms, NEO Gamma was designed specifically for operation inside private homes, with improvements spanning hardware reliability, noise reduction, locomotion, manipulation, and onboard artificial intelligence. 1X opened pre-orders for the production version of NEO (based on the Gamma design) on October 28, 2025, at an Early Access price of $20,000, with first deliveries to U.S. customers scheduled for 2026.[1][2]
NEO Gamma is notable for being one of the first humanoid robots explicitly engineered for the consumer home market rather than industrial or warehouse applications. Its design prioritizes psychological acceptability through soft materials, low mass, near-silent operation, and a non-threatening silhouette wrapped in a seamless knit exterior.[3]
1X Technologies was founded in 2014 as Halodi Robotics by Norwegian roboticist Bernt Oivind Bornich in Moss, Norway. The company initially focused on developing safe actuators and full-body control systems for robots capable of working alongside people. In 2018, it released its first humanoid robot, EVE, a wheeled platform designed for logistics, security, and healthcare environments. EVE was piloted across several industries, including a 140-unit agreement with ADT Security Services in 2022.[4]
In March 2023, the company rebranded as 1X Technologies and shifted its strategic focus toward the consumer market. That same month, it closed a $23.5 million Series A2 funding round led by the OpenAI Startup Fund, marking one of OpenAI's first major investments in physical robotics. In January 2024, 1X secured $100 million in Series B funding led by EQT Ventures, with participation from Samsung NEXT, the Nistad Group, and other investors. The company's total funding exceeded $125 million, with a post-money valuation of approximately $820 million.[5][6]
The NEO platform represents 1X's pivot from wheeled enterprise robots to bipedal humanoids built for the home. Development progressed through several stages:
1X published the NEO Gamma announcement on February 21, 2025, accompanied by a 37-second promotional film shot inside an interior designed to resemble a high-end residence. The film opens with NEO Gamma powering on, then progresses through a sequence of household vignettes. Documented activities in the reveal video and adjacent press materials include preparing and serving coffee, straightening a crooked picture frame, carrying a basket of laundry, vacuuming, wiping a window, retrieving groceries, and watering houseplants. The robot is depicted moving deliberately and quietly, with arm swings synchronized to its walking gait.[1][9][20]
The reveal generated unusually broad mainstream coverage. Design publications such as Dezeen, Yanko Design, and Designboom focused on the knit cosmetic and Scandinavian styling, while technology outlets including TechCrunch, The Robot Report, and Live Science emphasized the underlying actuation, AI, and home-deployment strategy. Several outlets compared NEO Gamma to fictional domestic robots, with Live Science framing it as a real-world approximation of the Rosie robot from The Jetsons.[3][8][9][21]
NEO Gamma was designed with the explicit goal of making a humanoid robot that people would feel comfortable having in their homes. 1X's design team, led by former BMW designer Per Selvaag, pursued what the company describes as "psychological acceptability." This means the robot's form factor, materials, noise profile, and movement patterns were all chosen to minimize any sense of threat or unease in human occupants.[3]
Selvaag, who worked as lead vehicle designer at BMW and chief designer at Peugeot before founding the California design studio Montaag, brought an automotive sensibility to the project. Rather than emphasizing exposed mechanical structure, the team treated NEO Gamma as a piece of soft furniture that happens to walk, aligning it visually with sofas and knit throws rather than industrial machinery.[3][22]
1X has described the project as moving toward a Scandi-minimalist appearance, with the robot intended to "complement living spaces rather than disrupt them."[9]
At 30 kilograms (66 pounds), NEO Gamma is significantly lighter than most bipedal humanoid robots. For comparison, Tesla's Optimus weighs approximately 73 kilograms, and the Boston Dynamics Atlas platform weighs around 89 kilograms. The low mass reduces the kinetic energy of any potential collision, improving inherent safety around people, pets, and furniture.[8]
One of NEO Gamma's most distinctive features is its head-to-toe soft body construction. The robot is encased in a custom 3D lattice polymer structure that provides cushioning on all external surfaces. Over this lattice, NEO wears a seamless knit bodysuit made from durable nylon, 3D-printed using a Japanese Shima Seiki whole-garment knitting machine. This seamless construction allows the fabric to conform to NEO's body without impeding its range of motion or creating the unsettling mechanical sounds typically associated with robots.[1][9]
The bodysuit is available in three colors: tan, gray, and dark brown. The Scandinavian-influenced aesthetic was a deliberate choice to help NEO blend into residential interiors rather than appearing as an industrial machine. All joints are designed with no pinch points, further reducing the risk of injury during close interaction with people.[2]
The outer knit also functions acoustically, dampening the high-frequency hiss that typically escapes from a robot's actuator housings. According to 1X, the combination of cable-driven joints, polymer lattice cushioning, and the textile shell produces the robot's measured 22 dB operating sound level, roughly comparable to a domestic refrigerator. The knit is thus a load-bearing element of the user experience rather than a purely cosmetic finish.[3][9]
1X's proprietary Tendon Drive system forms the core of NEO Gamma's actuation. Instead of conventional gearboxes or harmonic drives, the system uses high-torque-density brushless DC motors connected to cable-driven differential transmissions, mimicking the tendon-and-muscle arrangement found in the human musculoskeletal system. This approach produces several advantages over rigid gear-driven actuators:[10]
1X publicly claims roughly a decade of internal development on the Tendon Drive concept, dating to early Halodi Robotics work that predates the bipedal NEO line. Cable-driven joints are historically attractive in research robotics because they yield naturally compliant behavior, but they have been slow to reach production because of persistent issues such as tendon stretch under load and frictional wear at routing pulleys. 1X has stated that NEO Gamma represents the first deployment of its Tendon Drive in which these issues have been solved sufficiently for unsupervised household use, with the 2-million-cycle endurance figure quoted as the headline reliability target.[10][22]
Compared with the NEO Beta prototype, Gamma's actuation block is reported as roughly 10 times more reliable in time between failures, attributed both to mechanical refinements in the cable assemblies and to revised manufacturing tolerances at the joint level. The same actuator family produces NEO Gamma's claimed 70 kg lift capacity and 25 kg carry capacity, demonstrated in lifting tests where the 30 kg robot picks up loads more than twice its body weight.[8][14]
NEO Gamma introduced "Emotive Ear Rings," LED-based visual indicators on each side of the robot's head that provide real-time feedback about the robot's state. These light rings change color and pattern to communicate whether NEO is listening, processing a command, performing a task, or idle, enabling more intuitive interaction without requiring a screen or display.[1]
The ear rings replace the more conventional face-screen approach used by some competing humanoids. 1X has argued that an explicit display can read as cartoonish or surveillance-like in a home setting, while a peripheral light ring conveys attention and state in a way that is closer to how a person might notice another person's posture or gaze direction. Pattern transitions are timed to align with the in-house LLM-generated body language so that gesture, gait, and indicator behavior remain consistent across a multi-step task.[3][9]
| Category | Specification | Value |
|---|---|---|
| Physical | Height | 1.65 m (5 ft 5 in) |
| Weight | 30 kg (66 lbs) | |
| Exterior | Custom 3D lattice polymer with seamless nylon knit bodysuit | |
| Mobility | Total degrees of freedom | 75 |
| Degrees of freedom per hand | 22 (44 total across both hands) | |
| Max walking speed | 1.4 m/s (5.0 km/h / 3.1 mph) | |
| Max running speed | 6.2 m/s (22.3 km/h / 13.9 mph) | |
| Locomotion control frequency | 100 Hz | |
| Strength | Lift capacity | 70 kg (154 lbs) |
| Carry capacity | 25 kg (55 lbs) | |
| Arm payload | 8 kg (18 lbs) per arm | |
| Finger speed | Up to 8.0 m/s | |
| Power | Battery capacity | 842 Wh |
| Runtime | Approximately 4 hours (typical use) | |
| Charge rate | Approximately 6 minutes per hour of runtime | |
| Compute | Processor | 1X NEO Cortex (NVIDIA Jetson Thor) |
| AI compute | Up to 2,070 FP4 TFLOPS | |
| Memory | 128 GB LPDDR5X | |
| GPU architecture | NVIDIA Blackwell | |
| Audio | Microphones | 4 (front, back, left, right) with beamforming and echo cancellation |
| Speakers | 3 (1 chest for voice, 2 pelvis for bass/360-degree sound) | |
| Vision | Cameras | Dual 8.85 MP 90 Hz stereo fisheye |
| Noise | Operating noise level | 22 dB |
| Protection | Hands | IP68 |
| Body | IP44 | |
| Actuation | Type | Tendon-driven (1X proprietary Tendon Drive) |
NEO Gamma introduced substantial improvements across multiple dimensions compared to the NEO Beta prototype. The following table summarizes the key differences:
| Feature | NEO Beta (August 2024) | NEO Gamma (February 2025) |
|---|---|---|
| Hardware reliability | Baseline | 10x improvement over Beta |
| Operating noise | ~32 dB | ~22 dB (10 dB reduction) |
| Walking gait | Functional bipedal locomotion | Natural human gait with synchronized arm swings |
| Sitting/squatting | Limited | Full squat, sit in chairs, kneel, lie down, and recover |
| Locomotion control | Standard control loop | Whole-body controller at 100 Hz using reinforcement learning from human motion capture |
| Object manipulation | Basic grasping | Visual manipulation model generalizing to unseen objects and environments |
| Language interaction | Integration with external LLM | In-house LLM for natural conversation and body language |
| Exterior design | Early prototype housing | Seamless 3D-knitted nylon bodysuit with soft 3D lattice polymer underlayer |
| Communication indicators | None | Emotive Ear Rings (LED-based visual feedback) |
| Audio system | Basic microphone/speaker | 4-microphone beamforming array + 3-speaker system |
| Walking speed | Up to 4 km/h | Up to 5 km/h |
| Running speed | Up to 12 km/h | Up to 22.3 km/h |
| Carry capacity | 20 kg | 25 kg |
In aggregate, 1X has positioned NEO Gamma as the first NEO iteration deemed suitable for unsupervised household trials. The Beta units were primarily research devices used inside controlled environments, while Gamma was the first hardware revision shipped to a wider set of homes during 2025. The company has stated that virtually every external surface, actuator subassembly, and software pipeline was revisited between Beta and Gamma, even when the visible specification value (such as the 1.65 m height) remained the same.[1][8][14]
NEO Gamma is powered by Redwood AI, 1X's proprietary vision-language model that serves as the unified control system for the robot's autonomous behaviors. Redwood is a 160-million-parameter transformer model that handles locomotion, manipulation, and navigation as a single end-to-end system rather than treating each as a separate module.[11]
The model integrates pre-trained language embeddings, vision tokens from a pre-trained vision transformer, and proprioception embeddings derived from sequences of joint positions and applied forces. These inputs pass through additional transformer blocks to extract a latent representation, which is then decoded into robot actions using a diffusion policy.[11]
A distinguishing technical feature of Redwood is its "whole-body control" approach. Walking, arm movement, grasping, and spatial reasoning are all managed jointly by the same neural network, allowing NEO to coordinate its full body during tasks. For example, the robot 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. The system predicts not only arm and hand commands but also walking, pelvis pose, and torso orientation commands simultaneously.[11][14]
Redwood runs entirely on NEO's onboard NVIDIA Jetson Thor GPU at approximately 5 Hz, meaning the robot can function fully without a constant internet connection. The model was trained on a large dataset of teleoperated and autonomous episodes collected from both EVE and NEO robots in company offices and employee homes. Notably, Redwood learns from both successful and failed interactions, allowing the system to improve from any experience regardless of outcome.[11]
Observers have characterized Redwood as deliberately small for a foundation-class robotics model. The 160-million-parameter scale is dwarfed by general-purpose large language models, and 1X has framed this compactness as a feature: a smaller policy fits comfortably within the Jetson Thor's memory budget, leaves headroom for parallel perception and audio pipelines, and runs on battery without aggressive thermal throttling. The company has signaled that future Redwood revisions will grow primarily through richer datasets rather than by inflating raw parameter count.[11][23]
Redwood is also notable for the explicit role it gives to failed rollouts. Most policy-learning approaches filter out unsuccessful demonstrations, but Redwood treats failures as informative training signals, learning which motions tend to drop a glass or knock over a cup so that the robot avoids those trajectories in similar future contexts.[11][14]
1X also developed a separate World Model, built on a 14-billion-parameter generative video backbone. This model was first trained on web-scale video data, then adapted with 900 hours of egocentric human video, and finally fine-tuned on 70 hours of NEO-specific sensorimotor logs. An inverse dynamics model (IDM) bridges the gap between visual predictions and physical actions by predicting the action sequences needed to move from one video frame to the next.[12]
The World Model enables NEO to learn new tasks from video data, even attempting actions the robot has never been explicitly trained on. 1X demonstrated this capability with tasks such as operating a toilet seat, opening a sliding door, ironing a shirt, and brushing a person's hair, all without prior examples for those specific tasks in the training dataset.[12]
1X's approach to building autonomous behavior in NEO follows a teleoperation-first strategy. Rather than waiting for full autonomy before deploying the robot, 1X ships NEO with a human-in-the-loop system where remote operators can take control using VR equipment when the robot encounters a difficult situation. Each teleoperated session simultaneously generates training data that is used to improve the Redwood AI model.[13]
CEO Bernt Bornich has described this approach as pragmatic: "If we don't have your data, we can't make the product better." The philosophy is that real homes present far more variety and unpredictability than laboratory environments, so collecting data from actual household deployments accelerates the path to full autonomy faster than any simulation-only approach could.[13]
In addition to structured teleoperation data, NEO was trained on 400 hours of "random play" data, where the robot explored environments without specific task instructions. This unstructured exploration taught the system what 1X calls "common sense physics," helping NEO recover gracefully from unexpected situations like slipping on a surface or encountering an obstacle such as a pet walking across its path.[13]
1X describes the pipeline that links teleoperation rollouts, autonomous rollouts, World Model rollouts, and Redwood updates as the path from supervised demonstration to grounded autonomy. Each demonstration captured in an early-access household feeds the training corpus that informs the on-device Redwood policy, so a new task performed by a remote operator can, after retraining, become an autonomous skill across the fleet.[13][14]
NEO Gamma introduced a new in-house large language model developed by 1X, replacing earlier reliance on external LLM providers. This model enables natural conversational interaction, allowing users to issue voice commands, ask questions, and receive spoken responses. The LLM also drives NEO's body language, synchronizing gestures and posture with verbal communication to create more natural interactions.[1]
The decision to develop an in-house LLM rather than continue licensing access to a third party was framed by 1X as both a privacy and a latency choice. Running the conversational model locally avoids streaming raw audio from a home to a cloud provider and reduces round-trip delay between a spoken command and the robot's response. The in-house model is tightly coupled to Redwood AI: when a user asks NEO to perform a task, the LLM decomposes the request into a sequence of intents that Redwood can execute, and produces the appropriate spoken acknowledgement and gesture timing while the task is in progress.[1][11]
NEO Gamma's computing platform is the "1X NEO Cortex," built on NVIDIA's Jetson Thor system-on-chip. Jetson Thor is powered by the NVIDIA Blackwell GPU architecture and provides up to 2,070 FP4 TFLOPS of AI compute within a 130-watt power envelope. The chip includes 2,560 CUDA cores, 96 fifth-generation Tensor Cores, a 14-core Arm Neoverse-V3AE CPU, and 128 GB of LPDDR5X memory with approximately 273 GB/s bandwidth.[15]
This level of onboard compute allows NEO to run the Redwood AI model, process vision and audio inputs, and execute control commands locally without relying on cloud connectivity. Real-time perception, reasoning, and motor control all run directly on the robot, providing consistent responsiveness even in homes with poor or no internet access.
1X pursues a strategy that is unusual in the humanoid robotics industry: deploying to homes first and industrial settings second. Most competitors, including Tesla (Optimus), Figure AI (Figure 03), and Agility Robotics (Digit), have focused initial deployments on factories, warehouses, and logistics centers where the environment is more controlled and the economic case is more straightforward.[16]
1X's rationale for the home-first approach is that domestic environments provide a richer and more challenging training ground for general-purpose embodied AI. Homes contain enormous variety in objects, layouts, surfaces, lighting conditions, and human behaviors. By collecting data from hundreds and eventually thousands of real homes, 1X believes it can develop more robust and generalizable AI models than would be possible in the relatively homogeneous environments of factories and warehouses.[16]
The deployment of NEO Gamma and its successors has followed this timeline:
NEO Gamma is designed to handle a range of everyday household tasks, including:
At launch, many tasks are performed through a combination of autonomous behavior and remote teleoperation. 1X has stated that the ratio of autonomous-to-teleoperated actions will shift progressively toward full autonomy through over-the-air software updates as the Redwood AI model improves with additional real-world data.[13]
Because NEO Gamma's early-access program depends on human teleoperators who can see through the robot's cameras and operate its limbs, 1X has invested in user-facing controls that govern when remote access is permitted. According to TechCrunch's reporting on the home trial program, the robot includes mechanisms that allow the household to determine when a teleoperator may take control of its surroundings, and sensitive areas inside the home can be designated as off-limits. The LED Emotive Ear Rings provide a visual cue when the robot is being driven remotely versus when it is operating autonomously, which the company has framed as a transparency feature for both occupants and visitors.[17][22]
1X set the Early Access price for NEO at $20,000, making it one of the most affordably priced bipedal humanoid robots on the market. For context, most humanoid robots from competitors either have no published consumer price or are priced well above $50,000. The company also offers a subscription option at $499 per month with a 6-month minimum commitment, lowering the barrier to entry for households that prefer not to make a large upfront purchase.[2]
The pricing strategy reflects 1X's belief that volume deployment is essential for collecting the training data needed to improve autonomy. A lower price point drives wider adoption, which in turn generates more diverse data from more homes, which feeds back into a better-performing AI model.
In March 2025, 1X and NVIDIA announced a research collaboration focused on advancing humanoid robot autonomy. 1X created a dataset API for NVIDIA to access data collected from 1X offices and employee homes. The collaboration produced a demonstration for NVIDIA CEO Jensen Huang's GTC 2025 keynote, where NEO Gamma autonomously performed a dish-loading task using a policy built on NVIDIA's Isaac GR00T N1 foundation model for humanoid robots.[18]
NEO Gamma's compute platform, the Jetson Thor, is itself an NVIDIA product, further cementing the relationship between the two companies.
In December 2025, 1X announced a strategic partnership with EQT, one of the world's largest private equity firms (with approximately 267 billion euros in assets under management). The deal envisions the deployment of up to 10,000 NEO humanoid robots across EQT's portfolio of more than 300 companies between 2026 and 2030. Target use cases span manufacturing, warehousing, logistics, facility operations, and healthcare. Pilots are expected to begin in the United States in 2026, with expansion across Europe and Asia to follow.[19]
This partnership is significant because it provides 1X with a large-scale commercial revenue stream and deployment footprint beyond the consumer market, while the industrial data collected from EQT portfolio companies feeds back into the same AI training pipeline used for home deployments.
The relationship between 1X and OpenAI dates back to 2022, when the two organizations began collaborating on integrating large language models with physical robotic systems. OpenAI's Startup Fund led 1X's Series A2 round in March 2023. NEO integrates AI models for task decomposition, planning, and natural language interaction, though 1X has progressively developed its own in-house language model to reduce dependence on external providers.[5]
Safety is a central design principle for NEO Gamma, given its intended deployment in homes with families. Multiple layers of safety are built into the hardware and software:[1][9]
NEO Gamma's reveal drew commentary that was simultaneously enthusiastic about the consumer framing and skeptical about the gap between marketing footage and real autonomy. Several reporters noted that the demonstrations included substantial teleoperated control rather than purely autonomous execution, and that 1X had been candid about this in interviews. Engadget and TechCrunch both highlighted the trade-off: customers placing a $20,000 deposit were effectively agreeing to host a partially remote-operated robot in exchange for a path toward greater future autonomy through over-the-air updates.[13][17]
Design critics responded positively to the knit cosmetic, with Dezeen describing NEO Gamma as "dressed head-to-toe in beige knitwear" and contrasting it with the metallic aesthetics of competing humanoid robots. Yanko Design framed the knit as a deliberate attempt to make robots feel less like exhibits and more like furniture.[3][9][21]
Industry analysts framed the February 2025 reveal as an inflection point for the consumer humanoid category, in that it crystallized a recognizable design language for home robots. Subsequent humanoid announcements in 2025 and 2026 were frequently benchmarked against NEO Gamma on weight, noise, and price.[8][14]
NEO Gamma competes in a rapidly growing humanoid robotics market that was valued at approximately $2.92 billion in 2025, with projections to reach $15.26 billion by 2030. Key competitors include:
| Company | Robot | Approach | Weight | Target market |
|---|---|---|---|---|
| 1X Technologies | NEO (Gamma) | Tendon-driven, home-first | 30 kg | Consumer homes |
| Tesla | Optimus | Rigid actuators, factory-first | ~73 kg | Manufacturing, then consumer |
| Figure AI | Figure 03 | Industrial-first | ~60 kg | Warehousing, logistics |
| Agility Robotics | Digit | Purpose-built for logistics | ~65 kg | Warehousing, fulfillment |
| Boston Dynamics | Atlas (Electric) | Research and demonstration | ~89 kg | Research, industrial |
| Unitree | G1 | Low-cost, general-purpose | ~35 kg | Consumer, industrial |
1X differentiates itself through its lightweight design, tendon-driven actuation (producing quieter and more compliant movements), consumer-first strategy, and aggressive pricing at $20,000.[16]
1X manufactures NEO at facilities in Hayward, California and Moss, Norway. The company's production roadmap envisions scaling from thousands of units in 2025 (for testing) to tens of thousands in 2026, with a target of 100,000 units by 2027 and millions by 2028. In August 2025, Vikram Kothari, who previously managed SpaceX's supply chain, joined as Vice President of Operations to oversee this scale-up.[4]
In July 2025, 1X relocated its global headquarters from Norway to Palo Alto, California, consolidating employees into an 80,000-square-foot facility capable of seating 400 people. As of late 2025, the company employs over 600 people.[4]
NEO Gamma occupies a transitional position in 1X's product line. It was the first NEO revision that the company itself described as ready for a consumer home rather than a research environment, and the design choices that originated with Gamma, including the knit shell, the Emotive Ear Rings, the 22 dB acoustic target, and the on-device Redwood AI policy, were carried directly into the production NEO offered for pre-order in October 2025. In that sense, the production NEO can be read as a hardened iteration of Gamma rather than as a fundamentally new platform.[1][2]
The broader implication of NEO Gamma for the humanoid robotics field is that it provided an early, well-documented blueprint for what a consumer-targeted humanoid might look and behave like: lightweight, soft-skinned, near-silent, and designed to share a sofa or hallway with a person rather than to occupy a fenced workcell.[8][14][16]