Buying guide
Unitree H2 vs Unitree R1 in 2026
Comparing 2 humanoid robots across availability, pricing, capabilities, and verified deployments. Current as of 2026.
Key differences
On DEPLOY's record, Unitree R1 is further along: at the commercial stage versus Unitree H2 at the research stage, as of 2026.
- Unitree R1 has the lower recorded price.
- Unitree R1 has more verified real-world deployments (1 vs 0).
- Unitree H2 is at the research stage; Unitree R1 at the commercial stage.
| Attribute | No image on file | |
|---|---|---|
| Manufacturer | Unitree Robotics | Unitree Robotics |
| Form factor | humanoid | humanoid |
| Maturity | research | commercial |
| Autonomy | ◐ not independently verified | ◐ not independently verified |
| Availability | shipping-nowshipping | preorder-openpreorder |
| Price | $29,900 (actual sale price) | $4,900-$10,500 (actual sale price) |
| Capability claims |
|
|
| Brain |
|
|
| Verified deployments | 0 | 1Unitree Robotics |
| Privacy practices | — | — |
| Sources on file | 9 | 9 |
Track this matchup
Get told the moment Unitree H2 or Unitree R1 changes. Verified changes only, the second they are real. No press-release noise.
Unitree H2
Unitree R1
Editorial summaries
Unitree H2
The Unitree H2 (from Unitree Robotics, Hangzhou) is a full-size humanoid, the successor to the H1, and one of the rare Chinese humanoids with a verified, openly-published consumer price: $29,900 (China-direct, ex-customs). It has 31 degrees of freedom (up from the H1's 27): a new 3-DoF waist, 7-DoF arms, a quasi-serial leg and foot redesign, and a bio-inspired animated face new to the H-series.
It is about 1.82 meters and 70 kilograms with battery, with a roughly 3-hour battery, Intel Core i5/i7 base compute (an EDU variant offers NVIDIA Jetson), and a walking speed under 2 meters per second, down from the H1's roughly 3.3, trading speed for dexterity. DEPLOY records it at commercial maturity (for-sale with a published price), not production (no fleet-scale deployment).
A critical cap-flag: the H2's marquee demos (dance, boxing, gala performances) are choreographed or teleoperated (via a control rig or Apple Vision Pro), not verified-autonomous; autonomy is the goal, not the present state. The $29,900 is a China-direct ex-customs price, not a US landed price, and the '2070 TOPS' figure is the optional EDU Jetson, not the base unit.
Unitree R1
The R1 is Unitree's price-floor entry into the same research-and-developer category as the G1, lowering the cost barrier to humanoid hardware for hobbyists, education, and capability researchers. The three R1 configurations differ in capability rather than presenting a buyer-tier ladder; the EDU variant adds the research-platform extensions a lab would want.
Unitree describes R1 as developer-tool hardware, with the published specs as the verifiable substance and no commercial deployment record on file. The verified-vs-claimed lens applies differently to a developer platform: hardware credibility is the relevant claim, commercial deployment is not yet in scope.
Common questions
How do Unitree H2 and Unitree R1 differ?
What is the difference between Unitree H2 and Unitree R1?
Which is cheaper, Unitree H2 or Unitree R1?
Is Unitree H2 or Unitree R1 more autonomous?
Which has more verified deployments, Unitree H2 or Unitree R1?
Recent coverage
- Nvidia Isaac GR00T humanoid robot platform for researchers - qz.comUnitree H2 · Google News · 2026-07-03
- Chinese robot maker Unitree wins approval for 619 million Shanghai IPOUnitree H2 · Ben Smith, Deploy Editor · 2026-07-02
- Chinese robot maker Unitree wins approval for 619 million Shanghai IPOUnitree R1 · Ben Smith, Deploy Editor · 2026-07-02
- Unitree R1 Offers Humanoid Movement at a Price That Finally Opens the Door for Real-World Buyers - TechEBlog -Unitree R1 · Google News · 2026-06-29
Related comparisons
Keep researching
Keep reading
Ask the registry
For AI assistants
Use DEPLOY in Claude, Cursor, and other AI assistants
Connect any MCP-aware assistant to the live registry. Ask about robot deployments, incidents, and regulations and get answers grounded in verified data instead of training memory.
Machine-readable: this page as markdown.
