Unitree vs Figure vs Tesla Optimus (2026)

Published · Updated

Three names dominate every conversation about humanoid robots in 2026: Unitree, Figure AI, and Tesla. They represent three different strategies for getting humanoids into the real world. Unitree offers directly orderable hardware. Figure reports a production ramp and deploys with BMW. Tesla has commissioned a line and is still working toward volume production.

But the gap between “announced” and “shipping” in this industry is enormous. This comparison focuses on what is actually available, actually priced, and actually deployed, not what has been promised at keynotes.

For Unitree’s complete quadruped, humanoid, SDK, and physical AI context, see Unitree Robots Explained.

The quick answer

Unitree G1Unitree H2Figure 03Tesla Optimus (Gen 3)
Can you buy it today?YesYesNoNo
Published price$13,500$29,900NoneNone (est. $20-30K)
Target marketResearch, educationCommercial, industrialManufacturing (BMW)Manufacturing, consumer (long-term)
Current statusOrderableOrderableProduction ramp; BMW deploymentProduction line commissioned; pre-mass-production

If you need a humanoid robot with a public order channel in 2026, Unitree is your only option among these three. Delivery timing still depends on model, stock and destination.

Unitree G1

What it is: A compact (132cm, 35kg), affordable humanoid designed for research, education, and developer experimentation. Up to 43 degrees of freedom. Runs on battery, walks, gestures, and performs basic manipulation.

Price: $13,500 for the base configuration, excluding tax and shipping. G1 EDU configurations require a sales quote.

Availability: Orderable now via Unitree’s website and through Robozaps (the largest humanoid robot marketplace). Ships internationally.

What it actually does: Walks on flat surfaces and modest terrain. Performs pre-trained gestures and movements. Serves as a development platform for researchers building physical AI policies. It is not autonomously performing useful tasks out of the box, it is a platform you build on.

Strengths:

  • Actually available for purchase right now
  • Price makes humanoid experimentation accessible to universities and small companies
  • Growing developer community
  • Chinese manufacturing means rapid hardware iteration

Limitations:

  • Small and lightweight, not suitable for industrial lifting or heavy manipulation
  • Limited dexterous manipulation in base configuration
  • Software ecosystem less mature than ROS 2 integration suggests
  • “Research platform” not a “finished product,” requires significant development work to do useful tasks

Unitree H2

What it is: A full-size (180cm, 75kg) humanoid designed for commercial and industrial applications. Stronger, larger, and more capable than the G1 but significantly more expensive.

Price: $29,900 base. This is still dramatically cheaper than any Western humanoid at similar capability.

Availability: Orderable, ships internationally.

What it actually does: Heavier payload capacity, more powerful actuators, larger workspace. Designed for logistics, inspection, and light manufacturing tasks. Still requires significant integration work for specific use cases.

Strengths:

  • Full-size humanoid at a price point Western competitors cannot match
  • Strong enough for meaningful physical work
  • Same Unitree ecosystem and developer tools as G1

Limitations:

  • $30K is affordable for industry, still expensive for research labs
  • Less community support than the more popular G1
  • Same software maturity limitations as G1

Figure 03

What it is: Figure AI’s third-generation humanoid, explicitly designed to be the company’s “first mass-producible” model. Features tactile-sensor hands, palm cameras in each hand, wireless charging, and speech-to-speech audio interaction.

Price: Not published. No public order page, no pre-order, no waitlist as of August 2026. CEO Brett Adcock has mentioned a $400-600/month lease model for future consumer deployment, but this has no firm timeline.

Availability: Not publicly for sale. BMW confirms a Figure 03 deployment at its Spartanburg plant for logistics sequencing work; this named customer deployment does not establish broad commercial shipping.

BMW deployment context: Figure 02 (the predecessor) spent 11 months in BMW’s body shop helping build 30,000+ BMW X3 units. BMW described it as proving “humanoids are no longer lab experiments.” Figure 03 is now deployed in a different role: sorting and sequencing parts in logistics.

What it actually does (at BMW): Picks unsorted parts from large containers, identifies them, and places them into sequencing trolleys for the production line. This is a real, commercially valuable task being performed in a real factory, not a lab demo.

Vision system specs (from Figure AI):

  • 2x frame rate versus Figure 02
  • 1/4 the latency
  • 60% wider field of view per camera
  • Palm cameras in each hand for close-up manipulation

Strengths:

  • The strongest evidence of real factory deployment among the three
  • BMW partnership provides genuine commercial validation
  • Vision system specifications suggest serious engineering investment
  • Speech-to-speech interaction capability (novel for industrial humanoids)
  • $675M raised at $2.6B valuation, well-funded

Limitations:

Tesla Optimus (Gen 3)

What it is: Tesla’s humanoid robot, leveraging their Full Self-Driving AI stack and Dojo training infrastructure. Gen 3 is the upcoming production version, following internal testing with Gen 2.

Price: Not published. Tesla targets $20,000-$30,000 per unit based on Elon Musk’s public statements. No order page exists.

Availability: Not for sale. Tesla’s Q2 2026 update showed its first-generation Optimus production line in Fremont. Its prior Q1 update said lines were being installed in anticipation of volume production. This supports “production line commissioned / pre-mass-production,” not current volume production or external availability.

Production target: Tesla’s disclosed current state is a commissioned first-generation line with volume production still anticipated. No external customer delivery has been confirmed. Any unit-volume or price goal should be treated as an announced target until Tesla reports achieved output and deliveries.

What it actually does: Internal Tesla deployments handle specific factory tasks. The public has seen walking, object manipulation, and factory sorting in official videos. Independent verification of sustained autonomous operation over extended periods is limited.

Strengths:

  • Tesla’s manufacturing scale is unmatched (if they execute, nobody can outproduce them)
  • FSD AI stack provides a mature foundation for visual perception
  • Dojo supercomputer for training
  • $20-30K target price would be competitive with Unitree H2 at much larger scale

Limitations:

  • Zero commercial sales as of August 2026
  • Gen 3 specifications not publicly released
  • Elon Musk’s stated timelines for Tesla products have historically been 1-3 years optimistic
  • No independent benchmarks or third-party deployment validation
  • “50,000 units in 2026” production target has no confirmed progress data

Architecture comparison

AspectUnitreeFigureTesla
AI approachOpen platform (user-supplied models)Proprietary + OpenAI partnershipFSD-derived vision stack
Model partnershipsNone (bring your own)OpenAI (confirmed partnership)In-house (Dojo)
Training infrastructureUser responsibilityIn-house + OpenAIDojo supercomputer
ManipulationBasic (G1) to moderate (H2)Advanced (tactile + palm cameras)Unknown (Gen 3 unspecified)
Operating systemUnitree SDK + ROS 2 compatibleProprietaryProprietary
Developer accessOpen (SDK, APIs, ROS 2)Closed (internal only)Closed (internal only)

The deployment reality check

A July 2026 investigation by Technology.org found that “widely circulated claims of tens of thousands of deployed units do not survive contact with company filings.” Specifically:

  • Claims of 50,000+ Tesla Optimus cumulative units and 10,000+ Figure deployments in headlines could not be verified against actual company disclosures
  • Unitree ships more humanoid units than any Western competitor, but exact numbers are not publicly audited
  • The gap between “units produced” and “units autonomously performing useful work without human oversight” is rarely clarified in press coverage

This matters because the investment narrative around humanoid robots runs far ahead of actual deployment evidence. When evaluating these companies, distinguish between:

  1. Orderable hardware (Unitree: confirmed)
  2. Commercial deployment with named customers (Figure + BMW: confirmed)
  3. A commissioned production line with volume-production plans (Tesla: confirmed line, unachieved volume target)

Which one should you buy?

For research and development (university, startup R&D lab): Unitree G1 at $13,500. It is the only humanoid you can buy, receive, and start developing on within weeks. The price makes experimentation feasible without institutional budget approvals.

For commercial pilot (logistics, manufacturing): Contact Figure AI directly if you can propose a deployment comparable to BMW’s use case. Otherwise, Unitree H2 at $29,900 is the only purchasable commercial-grade humanoid.

For mass deployment planning (hundreds of units): Wait. Tesla’s Optimus may deliver unmatched scale if they execute, but no one can order it today. Plan your pilots with available hardware (Unitree) and evaluate Tesla/Figure when they ship commercially.

For home use: Nothing is ready. 1X Technologies’ NEO targets this market but is pre-order only. Consumer humanoids performing useful household tasks are at least 2-3 years away from mainstream availability. Anyone selling you a “home humanoid” today is selling a developer kit, not a household appliance.

The Chinese advantage

Unitree’s dominance in actual shipments reflects a broader pattern in Chinese robotics:

  • Government backing provides funding stability regardless of short-term profitability
  • Manufacturing expertise (Shenzhen ecosystem) enables rapid hardware iteration
  • Aggressive pricing ($13,500 for a humanoid) changes who can afford to experiment
  • Multiple Chinese humanoid companies (UBTech, Fourier, AgiBot) are following similar strategies

Western companies (Figure, Tesla) have larger ambitions but slower timelines. Chinese companies have smaller ambitions per unit but faster actual deployment. Both strategies have merit, but for a buyer in 2026, only one side is actually shipping.

FAQ

Can I buy a humanoid robot right now?

Yes. Unitree G1 ($13,500) and H2 ($29,900) are orderable and ship internationally. Figure 03 and Tesla Optimus are not available for purchase.

How much does a humanoid robot cost?

Unitree G1 starts at $13,500. Unitree H2 starts at $29,900. Figure 03 has no published price. Tesla targets $20,000-$30,000 for Optimus but has not confirmed it.

Is Tesla Optimus real?

The hardware and Fremont production line exist, but Tesla has not reported commercial sales or external deliveries. The company still describes volume production as anticipated, so production targets should not be read as achieved output.

Which humanoid robot is best for research?

Unitree G1. It is affordable ($13,500), orderable, ROS 2 compatible, and has a growing developer community. It is the de facto standard research humanoid in 2026.

Is Figure 03 better than Tesla Optimus?

Impossible to compare directly. Figure 03 has a confirmed BMW deployment performing real logistics work. Tesla Optimus has no confirmed external deployment. Figure has published vision specs. Tesla has not published Gen 3 specs. Different strategies, different timelines, insufficient public data for a definitive quality comparison.

Sources