CYBORGS OF EARTH
Cyborg Index · July 2026

Agibot A2 vs XPeng IRON

Both humanoids read on the same 8 axes and 40 sub-axes, using the published Index scores. Every difference below carries the public evidence behind it, and sub-axes with no public evidence count as 0 rather than being skipped.

The public record is still thin on at least one side of this pair, so treat the ordering as provisional rather than settled.

Overall standing

Capability score · 0–5 over 40 sub-axes
Agibot A2
Agibot (Zhiyuan Robotics)
1.55of 5
Rank
#2
Coverage
26/40
Proof
0.19
XPeng IRON
XPeng
0.47of 5
Rank
#12
Coverage
11/40
Proof
0.01
The short answer

Agibot A2 scores 1.08 higher than XPeng IRON, but that lead rests mostly on vendor claims and demos.

Only 12% of Agibot A2's score is carried by independently verified material — the rest is claimed or shown in demos. XPeng IRON has more unscored ground than the gap between them, so this could change next month.

Where Agibot A2 is ahead
  • Autonomy+2.00Rests on claims and demos
  • Safety+1.40Rests on claims and demos
  • Locomotion+1.40Rests on claims and demos
  • HRI+1.40Rests on claims and demos
Where XPeng IRON is ahead
  • Perception+0.60Rests on claims and demos
What would change it

19 sub-axes on the trailing side have no public evidence yet. If all of them scored full marks they would be worth 2.38 — against a gap of 1.08.

  • Locomotion · Speed enduranceup to +0.125
  • Perception · Spatial memoryup to +0.125
  • Autonomy · Instruction followingup to +0.125
  • HRI · Speech ioup to +0.125
  • Locomotion · Uneven terrainup to +0.125
There is more unscored ground than there is gap, so this result could turn over after the next research month.
How much we know
46%of sub-axes covered

We have public evidence for 46% of the 40 sub-axes across these two. Treat this as an early read.

Agibot A2 · 65%
XPeng IRON · 28%
Early read
The shape of each score
Agibot A21.55
LocomotionManipulationPerceptionAutonomyLearningHRIDeploymentSafety

26/40 sub-axes have evidence

Faint outline = XPeng IRON

XPeng IRON0.47
LocomotionManipulationPerceptionAutonomyLearningHRIDeploymentSafety

11/40 sub-axes have evidence

Faint outline = Agibot A2

Reading the wheels — clockwise from the top
  1. 1 Locomotion
  2. 2 Manipulation
  3. 3 Perception
  4. 4 Autonomy
  5. 5 Learning
  6. 6 HRI
  7. 7 Deployment
  8. 8 Safety

Each wheel has 40 spokes — five per axis. A longer spoke is a higher score, and the dashed rings mark 1 to 4. Hover or tab onto any spoke to read the axis, the sub-axis and the exact value. A missing spoke means there is no public evidence for that sub-axis yet, which is not the same as scoring zero. The faint outline on each wheel is the other humanoid, drawn on the same scale.

Every axis, every sub-axis
Autonomy2.00Safety1.40Locomotion1.40HRI1.40Deployment1.40Manipulation0.80Learning0.80Perception0.60
Agibot A2XPeng IRON
Shared 0–5 track · widest gap first
Autonomy
2.00 / 0.00◄ 2.00
Rests on claims and demosEvidenced cells · 4/5 vs 0/5
Unattended operation
3.00 / —◄ 3.00
Task horizon
3.00 / —◄ 3.00
Generalization
2.00 / —◄ 2.00
Instruction following
2.00 / —◄ 2.00
Recovery from failure
— / —
Safety
1.60 / 0.20◄ 1.40
Rests on claims and demosEvidenced cells · 4/5 vs 1/5
Functional safety
2.00 / 1.00◄ 1.00
Certifications
2.00 / —◄ 2.00
E stop arch
2.00 / —◄ 2.00
Incident transparency
2.00 / —◄ 2.00
Force impact limits
— / —
Locomotion
1.80 / 0.40◄ 1.40
Rests on claims and demosEvidenced cells · 3/5 vs 1/5
Flat ground walking
3.00 / 2.00◄ 1.00
Uneven terrain
3.00 / —◄ 3.00
Stairs
— / —
Disturbance recovery
— / —
Speed endurance
3.00 / —◄ 3.00
HRI
2.00 / 0.60◄ 1.40
Rests on claims and demosEvidenced cells · 4/5 vs 1/5
Speech io
2.00 / —◄ 2.00
Nl instruction
3.00 / —◄ 3.00
Intent recognition
— / —
Safe proximity
3.00 / —◄ 3.00
Social presentation
2.00 / 3.00► 1.00
Deployment
2.00 / 0.60◄ 1.40
Rests on claims and demosEvidenced cells · 4/5 vs 2/5
Named pilots
2.00 / 2.00
Field hours
— / —
Customer diversity
2.00 / 1.00◄ 1.00
Supportability
2.00 / —◄ 2.00
Commercial posture
4.00 / —◄ 4.00
Manipulation
1.40 / 0.60◄ 0.80
Rests on claims and demosEvidenced cells · 3/5 vs 2/5
Grasp variety
2.00 / 2.00
Bimanual
3.00 / 1.00◄ 2.00
Force control
2.00 / —◄ 2.00
In hand reorient
— / —
Tool use
— / —
Learning
1.20 / 0.40◄ 0.80
Rests on claims and demosEvidenced cells · 3/5 vs 1/5
Imitation learning
— / —
Sim to real
2.00 / —◄ 2.00
Foundation policy
— / 2.00► 2.00
Data flywheel
2.00 / —◄ 2.00
Transfer fine tune
2.00 / —◄ 2.00
Perception
0.40 / 1.00► 0.60
Rests on claims and demosEvidenced cells · 1/5 vs 3/5
Sensor stack
— / 2.00► 2.00
Scene understanding
— / 1.00► 1.00
Spatial memory
2.00 / —◄ 2.00
Object permanence
— / —
Semantic world model
— / 2.00► 2.00
Segment length and thickness are the gapNo public evidence yet
Which is better for what

Each row is decided by the axes that matter for that kind of work, read off the same Capability scores as the ranking. Rows only appear where both humanoids have enough public record to answer.

Manufacturing and warehouse work
Named pilots and field hours, backed by what the hands can actually do.
Agibot A2
Agibot A2 scores higher on Deployment and Manipulation — 1.80 against 0.60 for XPeng IRON.
Running without a human in the loop
Unattended operation and task horizon, backed by what the robot can perceive.
Agibot A2
Agibot A2 scores higher on Autonomy and Perception — 1.47 against 0.33 for XPeng IRON.
Working alongside people
Functional safety and certification, plus how it reads and answers people.
Agibot A2
Agibot A2 scores higher on Safety and HRI — 1.73 against 0.33 for XPeng IRON.
Research and developer use
How much of the stack is documented and reproducible.
Agibot A2
Agibot A2 scores higher on Learning and Perception — 0.93 against 0.60 for XPeng IRON.
How solid the record is on each side
Agibot A2
Sub-axes on the record
26 of 40 · 65%
Carried by verified material
12%
Rests on claims and demos
1.36
Best documented — Locomotion, Manipulation, Autonomy, Learning, HRI, Deployment, Safety
Nothing on the record — every axis has at least one read
XPeng IRON
Sub-axes on the record
11 of 40 · 28%
Carried by verified material
2%
Rests on claims and demos
0.46
Best documented — Perception
Nothing on the record — Autonomy

What this comparison cannot settle

  • Neither humanoid has a public read on 10 of the 40 sub-axes, so those cannot be compared at all.
  • 23 sub-axes are read on only one side. A one-sided cell is undecided, not a zero for the silent side.
  • Public evidence covers 46% of the sub-axes across the two. Treat the overall ordering as an early read.
  • Most of Agibot A2's score rests on vendor statements and demo footage rather than independently verified material (12% carried by verified evidence).
  • Most of XPeng IRON's score rests on vendor statements and demo footage rather than independently verified material (2% carried by verified evidence).
  • Demonstrations happen in different environments and product revisions are not always equivalent, so like-for-like uptime and throughput figures do not exist for most of the field.

Go deeper on either humanoid

Nearby comparisons

All comparisons →
Cyborg IndexJuly 2026 editionCapability scoremethodology