Sanctuary AI Phoenix Review: Specs, Price & AI

Sanctuary AI Phoenix Review: Quick Buyer Verdict

Phoenix should be evaluated as a dexterity and Physical AI platform first, and a humanoid product second. Sanctuary AI has deliberately concentrated on the parts of human work that are hardest to automate: manipulating varied objects, using touch, learning from human demonstrations and adapting behaviour when rigid programming is insufficient.

That makes Phoenix unusually compelling from a technology perspective. It also makes procurement less straightforward than the specification tables commonly published for commercial industrial robots.

Pros

• Strong focus on fine manipulation rather than humanoid appearance alone.
• Advanced tactile-sensing programme for touch-driven tasks.
• Hydraulic robotic-hand technology aimed at high force, speed and dexterity.
• Carbon architecture combines robot learning, reasoning and task execution.
• Human-in-the-loop teleoperation provides a practical route for collecting training data.
• Commercial experimentation predates much of the current humanoid market.
• Generation 8 improved cameras, telemetry, sensing and manufacturability.
• Sanctuary’s Physical AI has demonstrated industrially relevant performance on difficult manipulation tasks.

Cons

• No public current Phoenix price.
• No complete Generation 8 specification sheet for payload, runtime, dimensions or power.
• Generation 8 is wheeled rather than bipedal.
• No ordinary buy-now or standard distributor procurement model.
• Many published demonstrations involve teleoperation or data collection rather than unrestricted autonomy.
• Public evidence from Sanctuary’s newest production work uses conventional industrial robotic hardware, not necessarily Phoenix.
• The commercial roadmap changed materially in 2026, making older Phoenix descriptions easy to misinterpret.

Our recommendation: shortlist Phoenix when your interest is specifically in Sanctuary AI’s dexterity, tactile sensing or Physical AI approach and you are prepared for a structured enterprise engagement. Do not shortlist it simply because you need “a humanoid robot.” Start with the task, ask Sanctuary which embodiment it would actually deploy, and compare that proposal with humanoids and conventional automation. Review the Sanctuary AI Phoenix listing at Anton Robots before starting a commercial enquiry.

How Much Does Sanctuary AI Phoenix Cost in 2026?

Sanctuary AI does not publish a current list price for Phoenix. There is no verified standard 2026 base price that buyers should use for budgeting, and Phoenix is not presented through Sanctuary’s current website as a conventional configurable hardware product with a public checkout price.

This is an important distinction. A company can accept commercial enquiries for humanoid embodiments without offering a standard Phoenix unit for ordinary sale.

A realistic commercial project may include far more than the robot itself:

Do not invent a Phoenix price from old articles

Unlike some humanoids with published starting prices, Sanctuary has not provided a reliable public Phoenix purchase figure. A speculative estimate is therefore less useful than a qualified commercial proposal.

If price transparency is important to your shortlist, compare Phoenix with the models in our 2026 humanoid robot price guide.

For Phoenix specifically, ask for:

1. Proof-of-concept cost: the complete amount required to demonstrate the defined task.
2. Production deployment cost: everything required to meet target cycle time and success rate in the real facility.
3. Ongoing annual cost: software, retraining, support, maintenance and human supervision.

The cheapest robot is irrelevant if it cannot solve the task. Likewise, advanced AI is poor value if conventional automation can meet the requirement with less cost and risk.

Is Sanctuary AI Phoenix Actually for Sale?

Not as a normal off-the-shelf humanoid product.

As of September 2026, Sanctuary AI’s commercial site focuses on deploying production-ready Physical AI across existing and future robotic hardware. Its commercial enquiry form still allows prospective customers to express interest in a Humanoid Embodiment, but that is different from publishing a standard Phoenix configuration with a price, lead time and order button.

Anton Robots therefore classifies Phoenix as a limited-availability platform rather than an ordinary catalogue purchase.

This does not mean Sanctuary has abandoned humanoids. It means the purchasing question has changed from:

“How much does Phoenix cost?”

to:

“What Sanctuary AI system would you deploy to solve this task today, and what role would Phoenix or a future humanoid play?”

That question is much more useful for an industrial buyer.

What Is Sanctuary AI Phoenix?

Phoenix is Sanctuary AI’s human-scale general-purpose robot platform, originally developed as the physical embodiment for its Carbon AI control system.

Unlike humanoid programmes that placed early emphasis on dynamic walking, Sanctuary concentrated heavily on hands, touch and object manipulation. The underlying thesis is simple: most economically useful human work requires interaction with objects, tools and environments designed around the human hand.

Generation 6, unveiled in 2023, was publicly described as approximately 5 ft 7 in tall, 155 lb in weight, capable of carrying up to 55 lb and travelling at up to 3 mph.

Later generations changed substantially.

Generation 7 concentrated on uptime, range of motion, tactile sensing, visual acuity, reduced manufacturing cost and faster task automation.

Generation 8 moved to a wheeled mobile base and was optimised even more explicitly for high-quality behavioural-data capture, improved sensing and faster manufacturing.

What Phoenix is

• A human-scale mobile manipulation and Physical AI development platform.
• A system designed around sophisticated hands and object interaction.
• A tool for collecting high-fidelity behavioural data.
• An embodiment for testing autonomous policies and teleoperated work.
• A platform Sanctuary has iterated through multiple hardware generations.

What Phoenix is not

• It is not currently a standard retail humanoid.
• Current Generation 8 is not a walking biped.
• It is not publicly specified with a complete current runtime, payload and environmental-rating sheet.
• It is not autonomous in every demonstration you see online.
• It is not proof that every human task is ready for robotic automation.
• It is not necessarily the hardware Sanctuary would recommend for your application in 2026.

If you are comparing the category rather than Sanctuary specifically, browse current humanoid robots before narrowing your shortlist.

Sanctuary AI Phoenix Generations: Why the Version Matters

A large amount of online information about Phoenix mixes specifications from different generations. Buyers should avoid doing this.

The key purchasing rule

If a supplier or article lists “Sanctuary AI Phoenix specifications,” ask which generation those figures describe.

Do not combine Generation 6 dimensions, Generation 7 hand improvements, Generation 8 mobility and Sanctuary’s newer 2026 hand programme into one fictional robot.

Sanctuary AI Phoenix Specifications

The table below deliberately separates historical published figures from what Sanctuary has actually confirmed for Generation 8.

Why this specification uncertainty matters

A 25 kg payload figure sounds useful, but it tells a buyer little without knowing where the load is carried, arm reach, centre of gravity, duty cycle and how much payload remains available during dexterous manipulation.

Likewise, a top speed is almost irrelevant if the commercial value comes from inserting, sorting, assembling or manipulating objects.

For Phoenix, the most important missing procurement specifications are arguably not height or speed. They are:

• Validated manipulation payload at relevant reaches.
• Cycle time for the buyer’s actual task.
• Autonomous success rate.
• Human-intervention frequency.
• Battery and charging behaviour.
• Safety architecture and applicable certification.
• Environmental limits.
• Production support and service expectations.

Require these for the exact proposed embodiment rather than relying on a three-year-old specification graphic.

Wheeled Mobility: Why Phoenix Generation 8 Does Not Walk

Generation 8’s wheeled base is one of the most misunderstood Phoenix design decisions.

Sanctuary says the decision was informed by customer feedback that bipedal legs were too frail to support the strong torso required for useful, precise and safe work.

That makes Phoenix unusual in a market where bipedal locomotion often dominates marketing.

Advantages of the wheeled approach

• Greater stability during manipulation.
• Lower fall risk on flat industrial floors.
• Potentially simpler control and maintenance.
• Less energy spent balancing and walking.
• A stronger foundation for high-force upper-body work.
• Simpler data collection in structured facilities.

Trade-offs

• Cannot use ordinary stairs like a capable biped.
• Curbs, large thresholds and uneven terrain can restrict access.
• Less suitable for buildings where tasks span non-accessible levels.
• Does not provide the same physical generality as a human-like lower body.

The interesting point is that Sanctuary itself once argued that true generality ultimately benefits from legs. Generation 8 nevertheless prioritises what creates useful work today: a stable torso and capable hands.

For a buyer, that is not necessarily a weakness.

A robot working beside a conveyor, workstation or production cell may gain little business value from walking dynamically. Stability, manipulation reliability and uptime can matter far more.

Phoenix Hands, Tactile Sensing and Dexterous Manipulation

Phoenix’s strongest technical story is its hands.

Sanctuary has repeatedly argued that the ability to manipulate objects is more important to useful general-purpose work than spectacular locomotion. That philosophy has driven acquisitions, patents, tactile sensing, hydraulic actuation and extensive hand-development programmes.

21-degree-of-freedom hand development

Sanctuary has demonstrated a 21-DoF dexterous robotic hand capable of in-hand manipulation. The company links this work directly to Phoenix and Carbon, using richer manipulation behaviour to improve the data available for AI training.

The company has also reported hydraulic valve actuators exceeding two billion test cycles without signs of leakage or degradation.

These are meaningful engineering milestones, although they should not be confused with a complete lifecycle qualification of every Phoenix system.

Why hydraulics?

Sanctuary argues that miniaturised hydraulic actuation can provide high power density, strength, speed and controllability inside a human-scale hand.

The objective is not simply to close a gripper around a box. It is to support activities such as:

• Changing an object’s orientation within the fingers.
• Handling small or irregular components.
• Applying controlled force during insertion.
• Detecting slip.
• Manipulating objects when vision is partially blocked.
• Using tools and interfaces designed for human hands.

Tactile sensing matters

Vision alone cannot always determine whether an object is slipping, whether a connector has seated correctly or whether excessive force is being applied.

Sanctuary’s tactile sensors are intended to capture that information directly at the hand.

The company has demonstrated applications including blind picking and slip detection, while richer tactile data can also improve the policies being trained for future autonomous manipulation.

Do not mix the old and new hand specifications

Sanctuary’s 2026 roadmap describes a newer 17-DoF hydraulic hand under development with improvements in performance, reliability and serviceability.

That does not mean Phoenix suddenly has a fixed 17-DoF specification. It shows that Sanctuary’s hand architecture continues to evolve.

Buyers should ask:

1. Which hand or end effector is proposed?
2. How many active degrees of freedom does that exact version have?
3. What tactile sensors are included?
4. What force, payload and cycle-life limits apply?
5. Can a damaged hand or finger module be serviced rapidly?
6. What autonomous task success rate has been demonstrated with it?

What Is Carbon AI?

Carbon is the AI control architecture Sanctuary introduced alongside Phoenix.

In its original Phoenix documentation, Sanctuary described Carbon as combining several approaches rather than relying on one model architecture.

Published capabilities and design concepts have included:

• Translation of natural-language goals into real-world action.
• Reasoning, task planning and motion planning.
• Symbolic and logical reasoning.
• Large language model integration.
• Deep learning and reinforcement learning.
• Goal-seeking behaviour.
• Physics-based simulation.
• Human-in-the-loop supervision.
• Teleoperation.
• Fleet-management concepts.

The important idea is that the robot must do more than recognise an object.

A useful system needs to decide what the object means in the task, where to move, how to grasp it, what force to apply, how to detect failure and what to do next.

Carbon vs Sanctuary’s newer “Physical AI” positioning

Current 2026 Sanctuary marketing places much more emphasis on Physical AI and hardware-agnostic task policies.

Buyers should therefore avoid assuming that every new Sanctuary project will be sold as a discrete “Carbon software licence” running on Phoenix.

The useful commercial question is:

What intelligence stack, task policy and data pipeline are included in this deployment, and on which hardware?

Is Sanctuary AI Phoenix Autonomous or Teleoperated?

Both approaches are part of Sanctuary’s development model.

This needs to be understood carefully because robot videos are easy to misinterpret.

Sanctuary has stated that its systems are capable of autonomous work. It has also been unusually direct in explaining that many videos used for its “Robots Doing Stuff” demonstrations involve human teleoperation.

That is not merely a fallback.

Teleoperation is used to collect the human behavioural demonstrations required to train robot policies.

Why teleoperation is useful

A human operator can demonstrate:

• Where the hand should move.
• How fingers should contact an object.
• How much force should be applied.
• How to recover when an object shifts.
• How different sensory signals correspond to successful completion.

Those demonstrations become training data.

Over time, an autonomous policy can learn to reproduce useful behaviour without the operator executing every movement.

The buyer question that matters

Do not ask only:

“Can Phoenix do this task?”

Ask:

“What percentage of successful production cycles are autonomous, how often does a human intervene, and what happens when the policy encounters a condition outside its training distribution?”

For production automation, that distinction can determine the entire ROI.

Why Phoenix Is Also a Behavioural-Data Platform

Generation 7 and Generation 8 increasingly positioned Phoenix as a system for collecting high-quality human behavioural data.

Generation 7 improved uptime, visual acuity, tactile sensing and range of motion while reducing manufacturing complexity.

Generation 8 added improved depth and vision cameras, a wider and higher-resolution field of view, improved telemetry, a stronger sensor suite and better audio-video capture.

That reveals something important about Sanctuary’s strategy.

Phoenix is not only being trained to do work.

Phoenix is also infrastructure for creating the datasets required to train future robots.

Why data quality matters

A dexterous manipulation policy may need to understand:

• What the robot sees before contact.
• The position and velocity of the hand.
• How the fingers are configured.
• Where contact occurs.
• How force changes during the movement.
• Whether the object slips.
• Whether the task ultimately succeeds.

Poor data creates poor policies.

For buyers, however, the distinction between data collection and production automation is essential. A robot that is valuable as a research and training platform is not automatically ready to replace a production worker at the required throughput.

Sanctuary AI’s 2026 Strategy Shift: The Most Important Part of This Review

On 17 June 2026, Sanctuary AI announced a material change in how it brings its technology to market.

Rather than waiting for humanoid hardware to become mature enough for broad commercial deployment, Sanctuary is now applying its Physical AI to existing commercially available industrial robots.

The company describes this as a hardware-agnostic strategy that can create value immediately while developing the intelligence required for future humanoids.

Why Sanctuary changed direction

The commercial logic is strong.

A factory may already have:

• Reliable industrial arms.
• Known safety systems.
• Established maintenance teams.
• Proven controllers.
• Available spare parts.
• High mechanical uptime.

What the factory may lack is the ability to automate a highly variable or contact-rich task.

Instead of replacing all that mature hardware with a humanoid, Sanctuary can apply learned Physical AI to the existing automation stack.

The 99.5% result

Sanctuary reported a 99.5%+ task success rate at a 2.54-second cycle time on a complex wire-plugging task for a global Tier 1 automotive supplier.

The task involved manipulating flexible material and inserting it into a moving target on a conveyor.

This is one of Sanctuary’s most commercially important published results.

But there is a crucial distinction:

the result demonstrates Sanctuary’s Physical AI approach; it should not be presented as proof that a Phoenix humanoid achieved 99.5% production performance.

That nuance materially changes how a 2026 buyer should interpret Phoenix.

What this means if you want Phoenix

If you contact Sanctuary with a manufacturing problem, the best solution may now be:

• An existing industrial arm.
• A custom end effector.
• A learned Sanctuary task policy.
• A conventional cell with Physical AI.

rather than a Phoenix humanoid.

That is not necessarily bad for the buyer. In many cases it could be the faster, cheaper and lower-risk deployment.

Real-World Sanctuary AI Deployment Evidence

Sanctuary has more real-world history than a superficial reading of the humanoid market might suggest, but the evidence needs to be attributed correctly.

How strong is the Phoenix evidence?

The answer depends on what you are evaluating.

For dexterous AI research and data collection, Phoenix has a credible and technically differentiated history.

For a standard production humanoid product, the evidence is less mature because Sanctuary has deliberately widened its commercial approach beyond Phoenix.

That distinction should increase buyer confidence, not reduce it: the company is openly choosing the embodiment that can reach industrial performance sooner.

Best Sanctuary AI Phoenix Use Cases

1. Dexterous manipulation research

Best overall Phoenix use case. Organisations researching how robots grasp, reorient and manipulate varied objects can benefit from Sanctuary’s hands-first architecture and tactile-data approach.

2. Physical AI and embodied-AI development

Phoenix can act as an embodiment for collecting the multimodal behavioural data needed to train robot policies combining vision, touch and movement.

3. Industrial task-learning programmes

Factories with difficult, variable tasks may benefit from a Sanctuary engagement even when the final production solution does not use Phoenix hardware.

Typical candidates include:

• Connector and wire handling.
• Complex insertion.
• Sorting irregular parts.
• Flexible-material handling.
• Fine assembly.
• Tasks difficult to fixture perfectly.

4. Teleoperation and human-demonstration data

Phoenix is particularly relevant where the company needs humans to demonstrate varied task behaviour that can later be converted into autonomous policies.

5. Advanced robotics R&D partnerships

Universities, technology companies and large manufacturers investigating future general-purpose labour may find the Phoenix programme more relevant than organisations seeking a straightforward robot purchase.

6. Human-workstation manipulation

The human-like upper body and hand philosophy is valuable for objects, controls and tools originally designed for people.

However, the wheeled base means the environment must still support Phoenix mobility.

7. Strategic humanoid evaluation

For companies planning a multi-year humanoid programme, Phoenix represents a meaningfully different architecture from locomotion-first platforms and is therefore valuable in benchmarking.

For a broader shortlist, compare our guide to the best humanoid robots in 2026.

When Sanctuary AI Phoenix Is Not the Right Robot

Phoenix is technically compelling, but many buyers should choose something else.

• You need stairs: Generation 8’s wheeled base cannot provide the access of a capable biped.
• You need a robot this quarter: choose a platform with an explicit commercial delivery programme.
• You need a fixed budget: there is no standard public Phoenix purchase price.
• You need published battery endurance: current Generation 8 runtime is not publicly specified.
• You need outdoor or washdown operation: do not assume suitability without a written environmental rating.
• You need heavy transport: an AMR, forklift automation system or purpose-built material-handling robot may be more efficient.
• You need simple repetitive pick-and-place: a conventional industrial arm may deliver lower cost and higher throughput.
• You need home assistance: Phoenix is not a consumer domestic humanoid.
• You need an SDK-based research robot you can purchase freely: compare commercially available research humanoids.
• You mainly want a humanoid demonstration: other platforms have clearer purchasing channels and may meet that requirement with less integration work.

The most important reason to reject Phoenix is simple:

if the task does not require Sanctuary’s differentiated dexterity or Physical AI, you may be paying for the wrong capability.

Sanctuary AI Phoenix vs Figure 03, Digit, Apollo 2 and 1X NEO

These robots are often grouped together as humanoids, but their current commercial strategies are very different.

Which one should you choose?

• Choose or engage Sanctuary AI when dexterous Physical AI is the core technical problem and you are open to the best hardware embodiment for the task.
• Shortlist Figure 03 when you want one of the most vertically integrated AI-native biped programmes with rapidly scaling manufacturing.
• Choose Digit when the job is repetitive material movement in warehouses or factories and general-purpose hand dexterity is not required.
• Shortlist Apollo 2 for industrial humanoid partnerships where both wheeled and bipedal mobility may be useful.
• Choose NEO when the application is domestic rather than industrial.

Use the Anton Robots comparison tool to narrow the shortlist around your actual task rather than the robot’s appearance.

Is Sanctuary AI Phoenix Worth It?

Phoenix is worth evaluating when the economic bottleneck is difficult manipulation or task variability—and when Sanctuary can demonstrate a path to production performance that conventional automation cannot achieve economically.

It is poor value if the same task can be solved reliably using ordinary automation.

Start with the existing process

Calculate:

Annual opportunity = direct labour + overtime + unfilled-work cost + quality losses + downtime + ergonomic/safety burden − current automation cost.

Then compare that opportunity with the complete cost of the Sanctuary deployment.

Include all costs

• Proof-of-concept engineering.
• Robot or robots.
• End effectors and tooling.
• Physical AI development.
• Teleoperation and demonstration data.
• Fixtures and cell modifications.
• Safety integration.
• Networking and computing.
• Commissioning.
• Support and retraining.
• Internal engineering time.

Measure production outcomes

A serious pilot should establish:

• Success rate.
• Cycle time.
• Mean interventions per shift.
• Recovery time after failure.
• Quality or defect rate.
• Uptime.
• Changeover time.
• Human supervision required.

The June 2026 Sanctuary result is a good example of the right evaluation language: 99.5%+ success at 2.54 seconds per cycle is far more useful to a factory than saying an AI system is “general purpose.”

A practical go/no-go threshold

Do not approve a production deployment because the demonstration looks intelligent.

Approve it when the complete system meets your agreed success rate, cycle time, safety, intervention frequency and economic payback under representative conditions.

Sanctuary AI Phoenix Buying Checklist

1. Define the task. Specify the objects, starting conditions, goal state and required throughput.
2. Ask which embodiment Sanctuary recommends. Do not assume Phoenix is the production answer.
3. Identify the exact Phoenix generation. If Phoenix is proposed, require the hardware revision in writing.
4. Request current specifications. Payload, reach, dimensions, speed, runtime, charging, environmental limits and safety.
5. Define autonomy. Separate autonomous work, teleoperation, supervision and recovery.
6. Set the success-rate target. Define what counts as a successful cycle.
7. Set the cycle-time target. Average performance alone is insufficient; define allowable variation.
8. Measure interventions. Human rescue frequency can destroy the business case.
9. Validate the end effector. Confirm hand or gripper, tactile sensing, payload and service requirements.
10. Define the training process. Determine how demonstrations and production data become improved policies.
11. Review data governance. Document video, audio, tactile, telemetry and production-data handling.
12. Review safety. Evaluate the complete robot, tooling and surrounding equipment, not the AI in isolation.
13. Request lifecycle support. Know who maintains hardware and who retrains policies.
14. Calculate the three-year economics. Include internal engineering and continuing supervision.
15. Run a representative pilot. Use real parts, operators, disturbances and production conditions.

Pro tip: do not issue an RFQ that says “We want a Phoenix robot.” Issue one that says “We need this operation completed at X-second cycle time, Y% success and no more than Z human interventions.” Then let Sanctuary explain whether Phoenix, an industrial arm or another embodiment is the best solution.

How to Buy Sanctuary AI Phoenix

Phoenix is not currently purchased through a normal online hardware ordering process.

Sanctuary AI accepts commercial enquiries for automation projects and allows organisations to identify interest in industrial robotic arms, humanoid embodiments and robotic hands.

Before contacting Sanctuary or requesting a quote, prepare:

• A video of the existing task.
• Cycle-time requirement.
• Shift pattern and annual volume.
• Representative objects and product variability.
• Required forces and payload.
• Workspace dimensions and access.
• Existing automation and PLC interfaces.
• Safety requirements.
• Failure and quality criteria.
• Current labour and operating cost.
• Required deployment date.

Review the Sanctuary AI Phoenix product page for current marketplace status.

If Phoenix does not have the commercial availability your project requires, use Find My Robot or contact Anton Robots to compare alternatives with clearer delivery programmes.

What Is New for Sanctuary AI and Phoenix in 2026?

Physical AI moved ahead of the humanoid hardware

The biggest 2026 change is Sanctuary’s decision to deploy Physical AI on existing industrial robots rather than wait for humanoid hardware to become broadly commercial.

This arguably makes Sanctuary more commercially relevant in the short term, even though Phoenix itself becomes less straightforward as a product to buy.

99.5%+ industrial manipulation performance

Sanctuary reported more than 99.5% success at a 2.54-second cycle time on a challenging moving wire-insertion task.

This provides an industrial benchmark for Sanctuary’s intelligence stack rather than another staged humanoid demonstration.

Next-generation 17-DoF hydraulic hand

Sanctuary’s current hand-development roadmap describes a 17-DoF human-form hydraulic hand with improvements aimed at performance, reliability and serviceability.

It follows earlier 21-DoF hand work and shows that the hardware remains under active redesign rather than frozen into one permanent Phoenix specification.

Humanoids remain on the roadmap

Sanctuary’s current commercial language explicitly describes its Physical AI as supporting future robotic hardware and next-generation industrial humanoids.

Phoenix therefore remains strategically relevant.

The difference is that customers no longer need to wait for the complete humanoid form before using the underlying intelligence.

Sanctuary AI Phoenix FAQ

How much does Sanctuary AI Phoenix cost?

Sanctuary AI does not publish a standard 2026 Phoenix price. Current commercial engagements are project-based and may involve Phoenix, another robotic embodiment, Physical AI development, custom end effectors and integration. Request a complete project quote rather than relying on speculative online prices.

Can I buy Sanctuary AI Phoenix?

Phoenix is not currently presented as a standard off-the-shelf robot that buyers can configure and order online. Sanctuary accepts commercial enquiries including interest in humanoid embodiments, but the proposed solution may instead use existing industrial robotic hardware.

Is Sanctuary AI Phoenix available for sale in 2026?

Availability is limited and project-dependent. Treat Phoenix as an enterprise development and strategic deployment platform rather than a normal retail humanoid.

Which Phoenix generation is current?

Generation 8 is the latest publicly announced Phoenix generation. It uses a wheeled base and was designed around improved data capture, cameras, telemetry, sensors, human interaction and manufacturability.

Is Phoenix a bipedal robot?

The current Generation 8 Phoenix uses a wheeled base. Earlier Sanctuary design discussions considered bipedal mobility important to complete generality, but customer feedback led the company to prioritise a stronger, more stable torso on wheels for current work.

Why does Phoenix use wheels?

Sanctuary says customers indicated that bipedal legs could be too frail to support the strong torso needed for useful, precise and safe work. Wheels can increase stability and simplify operation on flat industrial surfaces.

How tall is Sanctuary AI Phoenix?

Sanctuary published a height of 5 ft 7 in, or approximately 170 cm, for Generation 6 Phoenix. The company has not publicly reconfirmed that figure as a complete Generation 8 procurement specification.

How much does Phoenix weigh?

Generation 6 was published at 155 lb, or approximately 70 kg. Do not assume Generation 8 has exactly the same weight.

How much can Phoenix carry?

Generation 6 was announced with a maximum payload of 55 lb, approximately 25 kg. Sanctuary has not published an equivalent detailed Generation 8 payload specification, so buyers should request limits for the exact proposed configuration.

How fast is Phoenix?

The 2023 Generation 6 announcement listed a maximum speed of 3 mph, approximately 1.34 m/s. Current Generation 8 mobility specifications should be confirmed directly because the platform changed to a wheeled base.

How long does the Phoenix battery last?

Sanctuary does not currently publish a clear Generation 8 battery-runtime specification. Any buyer evaluating Phoenix should request runtime, charging time, battery capacity and duty-cycle information for the exact system.

Does Phoenix have human-like hands?

Yes. Dexterous human-like manipulation is a core Sanctuary focus. The company has demonstrated 21-DoF hydraulic hand technology, tactile sensing and in-hand manipulation, although its hand architecture continues to evolve.

What is Sanctuary AI’s 17-DoF robotic hand?

Sanctuary’s 2026 development roadmap describes a newer 17-DoF hydraulic hand architecture intended to improve performance, reliability and serviceability. It should not be silently substituted for earlier Phoenix hand specifications; ask which hand is included in any proposed system.

Can Phoenix feel objects?

Sanctuary uses tactile sensing to capture information from physical contact. The company has demonstrated touch-driven capabilities such as slip detection and manipulation where vision alone is insufficient.

Is Phoenix fully autonomous?

Phoenix and Sanctuary’s broader technology can execute autonomous policies, but not every public demonstration is autonomous. Sanctuary has explicitly said that many demonstrations have used teleoperation to collect training data. Buyers should require autonomy and human-intervention rates for the exact task being proposed.

What is Carbon AI?

Carbon is Sanctuary’s AI control architecture developed for Phoenix. It combines perception, robot learning, reasoning, task and motion planning, simulation, reinforcement learning and human-in-the-loop methods to convert goals and demonstrations into physical behaviour.

What are Large Behavior Models?

Sanctuary uses the term to describe models trained on real-world behavioural information rather than text alone. Human demonstrations, robot telemetry, vision and tactile information can help an AI system learn how physical tasks are performed.

Did Phoenix work at Canadian Tire?

Sanctuary completed a week-long commercial deployment at a Mark’s store owned by Canadian Tire Corporation, reporting 110 retail-related tasks completed correctly. However, that January 2023 deployment used Sanctuary’s fifth-generation technology and occurred before Generation 6 Phoenix was unveiled. It should not be described as a Generation 8 Phoenix production deployment.

How many tasks can Sanctuary AI robots perform?

Sanctuary stated in its Microsoft collaboration announcement that its dexterous capabilities had been tested across 400 customer-defined tasks in 15 industries. That demonstrates breadth of experimentation, not 400 turnkey autonomous applications.

Does Sanctuary AI work with Microsoft?

Yes. Sanctuary announced a collaboration with Microsoft around AI models for general-purpose robots and the use of Microsoft Azure for training, inference, networking and storage.

Does Sanctuary AI work with Magna?

Yes. Sanctuary and Magna announced a strategic partnership involving general-purpose robot development and potential deployment in automotive manufacturing, along with work on cost and scalability.

What happened to Sanctuary AI’s humanoid strategy?

Sanctuary has not abandoned humanoids. In June 2026 it expanded its commercial strategy so Physical AI can be deployed on existing industrial robots now, while the same intelligence and dexterity work supports future humanoids.

What was Sanctuary AI’s 99.5% success-rate result?

The company reported more than 99.5% task success at a 2.54-second cycle time for a complex wire-plug insertion task at a global Tier 1 automotive supplier. The result supports Sanctuary’s Physical AI approach, but buyers should not attribute it directly to Phoenix hardware.

Is Phoenix better than Figure 03?

They prioritise different things. Phoenix is particularly interesting for dexterity, tactile sensing and Sanctuary’s hardware-agnostic Physical AI strategy. Figure 03 is a bipedal, vertically integrated humanoid with published five-hour runtime, active production scaling and substantial automotive deployment experience. The better choice depends on whether you need a complete humanoid platform or a solution to a specific automation problem.

Is Phoenix better than Agility Digit?

Phoenix aims at broader and finer manipulation. Digit has a narrower but clearer commercial focus on repetitive logistics and material-handling tasks. Choose Digit when warehouse movement is the requirement; investigate Sanctuary when manipulation complexity is the difficult part.

Is Sanctuary AI Phoenix worth it?

It can be worth a pilot when difficult manipulation or task variability prevents conventional automation from meeting production requirements. It is much harder to justify if the task can already be solved reliably with a standard industrial robot, AMR or purpose-built machine.

Final Verdict: Should You Buy Sanctuary AI Phoenix?

Do not evaluate Sanctuary AI Phoenix as if it were simply another humanoid robot for sale.

That misses what makes the programme important in 2026.

Phoenix has become a vehicle for Sanctuary’s work on dexterity, tactile sensing, behavioural data and Physical AI. Generation 8’s wheeled design reinforces that philosophy: useful manipulation and high-quality data have been prioritised over winning the race to make a robot walk like a person.

Sanctuary’s June 2026 strategy shift goes even further. The company is now willing to separate intelligence from embodiment and deploy its Physical AI on mature industrial hardware when that gets customers to production performance faster.

For a buyer, that leads to a clear conclusion.

Engage Sanctuary if your problem is difficult physical intelligence. Do not insist on Phoenix unless Phoenix is genuinely the best embodiment for the job.

Buyers seeking an immediately orderable humanoid with known runtime, delivery status and a stable hardware specification should compare other platforms.

Buyers tackling manipulation problems that conventional automation has failed to solve should take Sanctuary much more seriously.

The smartest evaluation is therefore task-first: define the production outcome, success rate, cycle time and human-intervention limit; let Sanctuary propose the embodiment; then compare that complete deployment with the best conventional and humanoid alternatives.

View the Sanctuary AI Phoenix at Anton Robots, browse other humanoid robots, or contact Anton Robots to compare the platform with commercially available alternatives.