Unitree G1 vs R1: price, hands and development access

As of October 2026, Unitree G1 starts at $13,500 and Unitree R1 at $5,900; R1 AIR starts at $4,900, before tax and shipping. The useful comparison is edition by edition: hands, sensors and development access differ. Unitree's Unitree G1 edition specifications and R1 table reserve secondary development for EDU.

Which G1 and R1 versions are being compared?

This comparison covers G1 base, G1 EDU, R1 AIR, R1 and R1 EDU from Unitree Robotics. Those are five edition columns in the maker's tables as of October 2026, but only two profiles in the atlas. An EDU label identifies a configuration family; it does not promise every optional part. See the G1 configurations and R1 configurations.

Simbeing's G1 and R1 comparison puts the model records side by side. The edition tables below add detail those records cannot express in a single height, price or joint-count field.

Read a specification as a statement about its named edition. Do not combine the lowest advertised price with the most equipped robot pictured nearby. A lab studying walking needs a different set of facts from a lab studying grasping. Neither can choose suitable hardware from a family name alone.

What does each dated price cover?

The published figures below are hardware starting prices in US dollars, excluding tax and shipping, as of October 2026. They are not a complete lab budget. Unitree supplies numerical base prices in its G1 product table and R1 product table.

The humanoid price and access-plan guide explains why a hardware listing is only one part of a project budget.

The G1 atlas record retains a historical $16,000 value in its structured price field. Its detailed price note already points to the newer $13,500 listing, confirmed by the maker's G1 store page as of October 2026. Using the old field alone would misstate this comparison.

Both store pages assign customs charges, tax and import clearance to the customer. They also separate additional equipment from the base listing. The R1 store page displays conflicting stock messages as of October 2026, so it does not support a firm delivery date here. A country selector also does not establish that a specific edition can be delivered to every listed destination.

For planning, keep the robot configuration, development equipment, transport costs, lab setup and staff time on separate lines. We have not estimated those extra costs. A published hardware figure is useful only when its limits stay attached.

How do size, joints and hands differ?

G1 is taller and heavier in the current desktop tables, while the joint count depends on edition. All figures below are maker-reported as of October 2026, from the Unitree G1 edition specifications and R1 table; they are not our measurements.

The detailed atlas entries are more useful than their broad hand tags: G1's component note says the base has no hands, while the R1 profile describes fixed standard hands. A fixed hand and an actuated gripping hand serve different research needs. Confirm the actual hand hardware before planning a manipulation task.

G1 also has a dimension conflict. Simbeing's source notes record maker pages giving 1,270 mm and 1,320 mm. This table uses the latter figure from the currently opened desktop table, as of October 2026. We have not established why the figures differ or averaged them into a new specification.

Load figures need the same care. The G1 developer guide reports about 2 kg arm load for base and 3 kg for EDU, with strong posture dependence. The Unitree R1 developer documentation labels about 2 kg as arm payload, but the R1 product page puts a kilogram value under a torque heading. As of October 2026, we retain that conflict and do not turn the atlas's missing R1 payload into a clean comparison. These figures do not establish total carrying capacity.

Which versions support secondary development?

Unitree's store notices explicitly restrict supported secondary development to EDU as of October 2026. That applies to both the G1 listing and the R1 listing. The product tables agree. General language about open interfaces should not override those edition-specific notices.

The G1 development guide separates the motion-control computer from the computer available to developers. The Unitree R1 developer documentation describes an EDU development backpack. Both guides cover joint control and service interfaces. Their scope does not make every interface available on every base robot.

Unitree also publishes SDK2 code and examples. That is evidence of available software, not proof that a particular hardware package supports it. Before relying on a sample, match its robot configuration, required accessories and software version to the proposed system.

For a lab, “can run custom code” is still too broad. Specify whether the project needs sensor readings, arm commands, joint-level control or an existing motion service. Then check the exact interface in the manual. This is a support question as well as a programming question; unofficial modification is outside this comparison.

What do the runtime and sensor figures mean?

The runtime figures describe the maker's estimates, not an identical-workload endurance test. As of October 2026, the G1 guide gives about 2 hours; the Unitree R1 developer documentation gives about 1 hour. The comparison does not establish how long either will run a lab's task.

These distinctions come from the linked developer guides. G1's guide describes an Orin NX development computer for EDU, while the current product table allows different high-compute modules. R1's guide describes optional Orin compute for EDU. Neither source supports assuming that all EDU configurations ship with the same computer.

Treat Simbeing's robot compute reference as background, then return to the actual hardware list. A processor family name does not specify the memory, power setting or software needed for an experiment. Likewise, a camera specification does not show how well a robot will handle reflective objects, changing light or clutter.

A useful lab trial should record the task, installed equipment, software and starting battery condition. Repeat the intended workload with the required logging and supervision. Those observations would answer a different question from this specification comparison: what the configured system does in your workspace.

Does either configuration establish household task readiness?

No: the evidence checked here does not establish reliable household cleaning or general autonomous chores. Unitree notes on its R1 product page that some displayed functions remain under development. A movement demonstration, hardware listing and supported household task are different forms of evidence.

For a chore-by-chore discussion, the home humanoid readiness guide distinguishes demonstrations, human assistance and household access.

For example, a grasping demonstration leaves questions about unfamiliar objects, interrupted tasks and human assistance. An EDU configuration may let a team work on those problems. That access is not itself a finished solution.

Describe the intended task before judging readiness. What object must move, from where to where? Who starts the task? What happens when the object is missing or falls? A useful evaluation records the help provided as carefully as the action completed. Without that evidence, this comparison can guide further research but cannot promise a household outcome.

Which questions should a lab resolve first?

Start with the task and required interfaces, then match them to an exact edition and equipment list. Use this checklist to make unknowns visible before building a project around a headline price.

1. Task: Is the project about walking, perception, grasping or human interaction? Define what successful completion looks like.
2. Edition: Which base or EDU configuration is documented? Record the hand and joint options separately from the model name.
3. Interfaces: Which commands and sensor streams are required? Find them in the relevant manual and confirm supported access.
4. Compute: Which development computer is included? Check memory, storage, power and software needs against that hardware.
5. Workspace: How will the robot be handled, supervised and stopped? Work from the maker's operating instructions for the actual configuration.
6. Support: Which software versions, accessories and operating conditions does the documented support cover?
7. Evidence: Which claims come from specifications, demonstrations or your own trials? Keep those categories separate when reporting results.

This process may narrow the choice, or show that more evidence is needed. A useful comparison can end with an unanswered question; it does not need a universal winner.

How we checked this

We checked the G1 and R1 records in public/data/humanoids.json at repository revision 7bce4d9bffb3dab3e37d164920b0b5d24cb1095c, then opened Unitree's product tables, store pages, developer guides and SDK repository on October 11, 2026. The edition matrix is our reconciliation of those sources. We did not test either robot.

We preserved the historical price and height conflicts, separated optional hardware, and kept the inconsistent R1 load labels visible. Missing values remain unknown. These are October checks for a planned later article; prices, configurations and support need a fresh check before publication.

Statistics only. No affiliation, sponsorship or endorsement.