Somnia Lab Model L Explained: The AI Companion Robot Behind the Viral “165 Positions” Claim

Key Takeaways

• Somnia Lab is developing Model L as an embodied AI companion and intimacy robot, combining robotics, computer vision, touch and force sensing, voice interaction, personas, memory, and body heating.
• The viral “165 positions” figure comes from an earlier prototype-era claim. Somnia's current Model L page instead advertises 100+ interactions, with additional positions planned through OTA updates.
• The widely repeated 175 cm / 20 kg / 24-DOF specification describes an earlier prototype, not the current Model L lineup. Current versions range from 22 kg to 55 kg and from 26 to 89 combined head-and-body degrees of freedom.
• Model L is not designed to walk. Somnia is deliberately focusing its first generation on controlled physical interaction, sensing, warmth, voice, memory, and body responsiveness rather than general-purpose humanoid locomotion.
• Pre-orders are not publicly open yet. Somnia currently says “Pre-orders open soon” and directs potential buyers to a waitlist. Final retail pricing has not been announced.
• Somnia has publicly announced a $10 million angel round, although the investors were not identified in the announcement.
• The biggest questions are no longer simply whether the concept exists. The important issues are production readiness, physical safety, privacy, durability, pricing, and whether Somnia can manufacture and support Model L at consumer scale.

What Is Somnia Lab?

Somnia Lab is an AI hardware and robotics company focused on what it calls intimacy robotics.

Its flagship product, Model L, is not positioned as a general household humanoid. Instead, Somnia is narrowing the robotics problem to a controlled environment where conversation, physical presence, movement, touch sensing, warmth, roleplay, and long-term memory can be integrated into one system.

That makes Model L relevant beyond the adult-product category. From an AI industry perspective, it is an example of a broader trend toward embodied AI companions: AI systems that have a persistent physical body and connect software intelligence to sensors, actuators, memory, and real-world interaction.

The current Model L concept combines:

• Humanoid robotic movement
• Binocular vision
• Touch and force sensing
• Body heating
• Voice and app control
• AI personas
• Scenario and roleplay systems
• Persistent memory
• OTA-expandable motion capabilities

Somnia's central bet is straightforward: the first commercially successful consumer humanoid may not need to clean a house, carry groceries, or walk through a city. It may succeed by performing one emotionally valuable job extremely well.

Is Somnia Lab Real?

Yes, but it is important to distinguish between a real company, a real prototype, and a production-ready consumer product.

Hong Kong 01 published an on-site report in August 2026 describing Somnia's development work in Shenzhen and an early humanoid prototype. Founder and CEO Xara, identified by the publication as using a pseudonym, discussed the robot's structure, movement system, sensing, AI behavior, funding, and commercialization plans.

That gives the project stronger evidence than a startup represented only by CGI renders or concept videos.

However, many of Model L's most ambitious capabilities remain company-provided specifications rather than independently benchmarked results.

The current evidence can therefore be separated into four levels:

• A physical prototype has been reported by journalists.
• Somnia has published detailed Model L specifications.
• The company has announced funding and a commercialization roadmap.
• Independent long-duration testing of production hardware is not yet publicly available.

That does not mean the specifications are false. It means they should be attributed correctly rather than presented as independently verified laboratory results.

Where Did the “165 Positions” Claim Come From?

The 165 figure is what pushed Somnia Lab into viral technology coverage.

In the August 2026 Hong Kong 01 interview, Xara described an approximately 1.75-meter, 20-kilogram prototype with 24 degrees of freedom and said the team had used human motion capture to build a library covering 165 intimate movements.

The important word is prototype.

A motion-capture library is not the same thing as a list of fully validated retail behaviors. Human motion still needs to be converted into robot-safe trajectories that respect constraints including:

• Joint limits
• Motor torque
• Structural loading
• Collision avoidance
• Contact force
• User position
• Thermal limits
• Mechanical stability

Somnia's current Model L product page no longer presents “165 positions” as the main specification. Instead, it advertises 100+ interactions and says more positions will roll out over time.

The most accurate interpretation is therefore:

165 refers to an earlier prototype-era motion-capture and development claim, while 100+ interactions is Somnia's current public Model L product language.

This distinction matters because viral coverage often treats the two claims as interchangeable.

Current Model L Specs

Somnia now presents Model L as a four-tier product family.

Somnia also lists binocular cameras, a built-in speaker, touch and force sensing, heated detachable modules, and OTA-expandable movement capabilities.

The table reveals how much the product has evolved since the first viral prototype coverage.

The earlier prototype was described approximately as:

175 cm + 20 kg + 24 DOF

The current Model L lineup ranges from:

165 cm + 22 kg + 26 combined DOF

up to:

175 cm + 55 kg + 89 combined DOF

The 175 cm Founder Edition is therefore almost three times the weight associated with the original 20 kg prototype claim.

That increase is not necessarily a negative sign. Real robots become heavy quickly when developers add stronger actuators, batteries, gearboxes, sensors, thermal systems, structural reinforcement, facial mechanisms, wiring, controllers, and safety margins.

For a physical robot, the useful question is not simply whether it is lightweight. It is whether it can deliver repeatable movement, controlled force, low noise, acceptable battery life, and safe interaction at its final weight.

Why Model L Does Not Walk

One of the most important decisions Somnia has made is what Model L does not attempt to do.

The official product information says Model L does not walk and is designed around bedroom use. Somnia argues that walking and general-purpose dexterous manipulation would add substantial weight, noise, power consumption, cost, and mechanical complexity.

This is strategically significant.

General-purpose humanoid developers need to solve problems including:

• Dynamic balance
• Navigation
• Stairs
• Uneven floors
• Falls
• Object manipulation
• Doors and furniture
• Whole-body coordination
• Highly variable environments

Somnia can avoid much of this problem space.

Instead, engineering resources can be concentrated on:

• Controlled close-range movement
• Touch sensing
• Force response
• Thermal comfort
• Voice interaction
• Persistent memory
• Quiet operation
• Physical stability

This does not make Model L easy to build. Safe close-contact robotics is itself a difficult engineering problem.

But it makes Somnia's strategy easier to understand: reduce the scope of humanoid robotics until the remaining product solves a narrower problem that some users may be willing to pay a premium for.

Motion Capture May Matter More Than the Number of Positions

The most technically interesting part of Somnia's approach is not whether the motion count is 100, 165, or eventually something higher.

It is the use of human motion capture as reference data for robotic behavior.

A simplified motion pipeline could look like this:

Human motion capture
        ↓
Pose reconstruction
        ↓
Robot kinematic retargeting
        ↓
Joint and collision constraints
        ↓
Force and contact constraints
        ↓
User-body adaptation
        ↓
Real-time motor control

The difficult stage is kinematic retargeting.

A human body and Model L do not have identical joint locations, bone lengths, mass distribution, flexibility, balance, or mechanical compliance. A recorded human movement therefore cannot simply be copied directly onto the robot.

The control system has to translate the motion into something the robot can physically execute.

That involves solving for factors such as:

• Joint-angle limits
• Torque limits
• Collision constraints
• Stability
• Contact force
• User position
• Actuator speed
• Thermal conditions

This is why a large motion library is only the beginning.

The more meaningful product metrics will eventually be:

• Transition smoothness
• Position accuracy
• Contact awareness
• Real-time correction
• Force limiting
• Repeatability
• Failure recovery

Those metrics will become far more important than the headline number once independent reviewers can access production hardware.

How Model L Uses AI

Model L is not presented as a conventional robot with a chatbot connected to a speaker.

Somnia's product architecture connects AI to multiple layers of physical context.

Vision

All four Model L tiers currently list binocular cameras.

Two-camera vision can potentially help with relative positioning and depth estimation, although Somnia has not published independent benchmarks for perception accuracy or latency.

Touch and Force Sensing

Somnia lists touch and force sensing across multiple areas, with more extensive coverage on higher-end models.

This matters because an embodied companion needs to react to actual physical contact rather than simply replay a fixed animation.

Intent Recognition

Somnia lists intent recognition on Model L Pro, Pro Max, and Founder Edition, but not on Basic.

The company has not yet publicly documented enough of the technical stack to determine exactly how that system works, how much processing happens locally, or what accuracy users should expect.

Persona and Roleplay

Somnia separates physical appearance from software personality. The product page describes personas and scripts that determine the robot's character and interaction context.

That is an important product-design choice because the same physical platform could support multiple software-defined experiences.

Persistent Memory

Model L is also intended to remember information and previous interactions. Somnia says users will be able to view, change, or erase stored memories.

Conceptually, the architecture could look something like this:

Voice + vision + touch + force
            ↓
      Context / state model
            ↓
 Persona + memory + intent
            ↓
 Dialogue + motion planning
            ↓
 Speech + physical response

The difficult problem is synchronization.

An embodied AI needs to understand not only what to say, but also what its body is doing, where the user is positioned, whether contact is occurring, and what physical action is safe to execute next.

That is fundamentally more complicated than operating an AI companion entirely inside a phone or browser.

Memory Could Become a Major Part of Somnia's Business Model

Somnia says more advanced intelligence and long-term memory will require cloud services, with an optional subscription planned for users who want those features. Subscription pricing has not yet been announced.

That means the business model could extend far beyond a one-time hardware sale.

Potential revenue layers include:

• Robot hardware
• Premium configurations
• Custom designs
• Hardware upgrades
• Cloud AI
• Long-term memory
• Personas
• Scripts and scenarios
• OTA movement updates
• Maintenance and repair
• Physical experience locations

The most interesting possibility is that the persistent AI identity eventually becomes more valuable than the physical body.

If a user can preserve the same persona and memories while upgrading hardware over time, the relationship data becomes the continuity layer between device generations.

Somnia itself says Model L is intended for long-term ownership and describes an upgrade philosophy built around improving the hardware without replacing the relationship.

That could create unusually high retention.

It also makes privacy one of the company's most important challenges.

Privacy Is a Core Product Problem

Model L combines some of the most sensitive consumer sensors and data types imaginable:

• Cameras
• Microphones
• Physical-contact sensors
• Personal preferences
• Long-term memories
• Interaction history

The product is also intended for use in one of the most private environments in a person's home.

Somnia says users will be able to inspect, modify, or erase memories and says it plans to publish additional information covering camera and microphone processing, data storage, cloud services, and deletion.

Those details will be critical.

Important unanswered questions include:

• Does raw camera footage ever leave the device?
• Which computer-vision tasks run locally?
• Is audio continuously processed?
• What information is retained by default?
• Are memory summaries or embeddings stored remotely?
• Can users completely disable cloud processing?
• Which features continue to work offline?
• How are stored memories encrypted?
• What happens when a Model L is resold?
• Can users export their memory data?
• What happens to cloud-dependent features if Somnia shuts down?

For an embodied AI companion, privacy is part of the core product architecture, not merely a legal-policy requirement.

Somnia Lab Model L Price

Somnia has not announced final Model L retail prices.

The official waitlist provides useful evidence about the company's current pricing stage.

It asks potential buyers what price range they would consider:

• Under $5,000
• $5,000–$10,000
• $10,000–$20,000
• More than $20,000

The waitlist also asks which Model L configuration interests the user most.

That strongly suggests Somnia is still performing price discovery rather than marketing a finalized MSRP.

Older interviews have mentioned pricing ambitions, but those numbers should not be presented as the current official price.

As of September 25, 2026, the accurate answer is:

Model L's final retail price has not been announced.

Can You Pre-Order Model L?

Not through the main public product flow yet.

Somnia currently states “Pre-orders open soon” and directs interested customers to its Model L waitlist.

Current status:

This matters because older reports and screenshots may refer to earlier reservation experiments or deposit flows.

For current availability, Somnia's live public product pages should take priority over older material.

When Will Model L Ship?

Hong Kong 01's August 2026 interview reported that Somnia was targeting 2027 for commercial delivery.

However, the current Model L product page does not provide a firm guaranteed shipping date. Instead, it emphasizes the waitlist and upcoming reservations.

That makes 2027 best understood as a company roadmap target, not a guaranteed delivery date.

Shipping a humanoid consumer robot requires considerably more than building a working prototype.

A simplified development path looks like this:

Prototype
   ↓
Engineering validation
   ↓
Design validation
   ↓
Production validation
   ↓
Safety and regulatory work
   ↓
Pilot manufacturing
   ↓
Mass production
   ↓
Customer delivery

For Model L, the validation requirements are unusually demanding because users will interact physically with the machine.

Production hardware will need to address issues including:

• Actuator faults
• Joint wear
• Battery safety
• Thermal management
• Force limits
• Pinch hazards
• Material durability
• Cleaning
• Emergency stopping
• Sensor failures
• Software recovery

A successful prototype is therefore an important milestone, but it is not the final milestone.

Somnia Lab Funding

Somnia Lab publicly announced a $10 million angel round in 2026.

The company said the capital would support hiring and research across embodied AI, robotics, human-machine interaction, perception, motion, personalization, and safety.

The announcement did not identify the investors.

Separately, Xara has described Somnia as having attracted eight-figure-dollar funding in media interviews.

Those claims should not automatically be treated as identical evidence.

The strongest public statement is:

Somnia Lab says it completed a $10 million angel round.

Until more investor names, filings, or detailed financing announcements become public, broader cumulative funding claims should remain attributed to the company or founder.

Why Somnia Lab Matters for Embodied AI

Somnia is interesting even if its final market turns out to be smaller than the company expects.

Its strategy illustrates an important question facing the entire humanoid robotics industry:

Does the first widely adopted consumer humanoid actually need to be general purpose?

Many humanoid companies are pursuing factories, warehouses, household labor, and general manipulation.

Those markets are enormous, but the robots must perform a very wide range of tasks with extremely high reliability.

Somnia is taking a different path:

Narrow use case
      ↓
High willingness to pay
      ↓
Controlled environment
      ↓
Premium early adopters
      ↓
Usage data
      ↓
Better hardware and software

That is a familiar startup commercialization strategy.

Instead of solving every robotics problem before selling anything, a company identifies a specific environment where today's imperfect hardware may already provide enough value to justify the cost.

Model L can therefore be understood as a case study in vertical embodied AI.

The same pattern could eventually appear in other categories including:

• Elder companionship
• Therapy support
• Education
• Entertainment
• Hospitality
• Fitness coaching
• Specialized home assistance

The broader implication is that consumer humanoids may first succeed as vertical products before becoming universal household robots.

What Would Prove Model L Is Ready for the Market?

Somnia has already demonstrated that it can attract attention.

The next stage requires different evidence.

The strongest proof would be long, minimally edited demonstrations of production-intent hardware showing:

• Startup and calibration
• Voice-response latency
• Vision tracking
• Multiple movement transitions
• User-position adaptation
• Force-response behavior
• Emergency stopping
• Actuator noise
• Battery endurance
• Thermal stability
• Cleaning and maintenance
• Offline behavior
• Memory controls
• Recovery from interrupted movement

A polished 30-second video can demonstrate a compelling concept.

A continuous 30-minute demonstration can reveal whether the system is becoming a dependable product.

Independent reviewers should eventually focus on those same areas rather than simply counting the number of advertised movements.

Verified Facts vs. Forward-Looking Claims

The safest way to evaluate Somnia Lab is to separate current evidence from future promises.

This distinction is important for Google E-E-A-T as well as for readers.

Company specifications should be attributed to the company. Prototype demonstrations should not be presented as production validation. Future plans should not be written as completed milestones.

Conclusion

Somnia Lab Model L is more interesting than the viral “165 positions” headline suggests.

The deeper story is a startup attempting to turn AI companionship into embodied AI hardware by deliberately narrowing the humanoid-robot problem.

Model L does not walk. Instead, Somnia is concentrating its first generation on controlled movement, touch and force sensing, body warmth, vision, conversation, personas, persistent memory, and close-range human interaction.

Several early claims also need updating.

The familiar 175 cm / 20 kg / 24-DOF specification belongs to an earlier prototype description. Somnia's current lineup ranges from 22 kg to 55 kg and from 26 to 89 combined degrees of freedom. The viral “165 positions” language has also evolved into the current official claim of 100+ interactions with additional positions planned through OTA updates.

Most importantly, Model L is not yet a normal retail product. The waitlist is open, public pre-orders are still listed as coming soon, and final retail pricing has not been announced.

For anyone tracking AI hardware, physical AI, companion robots, or embodied agents, Somnia is worth watching because it represents a different commercialization path from general-purpose humanoid robotics.

The next signals will matter far more than the viral headline: final pricing, open pre-orders, production-intent hardware, independent testing, privacy architecture, manufacturing readiness, and real customer deliveries.