Humanoid robot rankings
A monthly ranking based on what each robot has shown publicly — and how strong the evidence is.
- Rank
- The public record across the same 40 checks.
- Evidence
- Claims and demos count less than independent or real-world results.
- Unknown
- No public evidence yet does not mean the robot is incapable.
ScopeThis ranks the public record, not undisclosed internal capability.
- moved up2
- moved down7
- held3
- new1
UBTech Walker S
Public evidence for 30 of 40 checks
Cohort shape · medians and spread
UBTech Walker S ranks first in the August 2026 Cyborg Index, followed by Agibot A2 and 1X NEO. The ranking assesses thirteen humanoids across forty capability measures and discounts claims that lack independent or operational evidence.
Narrower cuts of the same edition: best humanoid robots · Chinese humanoid robots · how the ranking changed
What this ranking does and does not mean
- Rank is the public evidence record, not the "best robot," the most funded, or the most advanced in secret.
- Scores are out of 5. A top score around 1.5 means the whole field is still early — not that the leader is almost ready.
- Missing evidence counts as 0, not a verdict that the robot cannot do the thing.
UBTech Walker S
- Capability scoreFilled magenta arc — score the public record actually supports today. Height = 0..5.
- Proof scoreDashed graphite arc — analyst read from source triangulation. Sits over the same wedge for direct comparison.
- Not evidencedEmpty wedge silhouette. No arc drawn — we never fake a zero for a gap in the public record.
- Axis groupHeavier spoke every 5 wedges marks one of the 8 axes. Axis label + n/5 coverage sits just outside the outer ring.
- 0.19→1.60+1.41
- 0.13→1.60+1.47
- 0.09→1.60+1.51
- 0.06→1.80+1.74
- 0.05→0.40+0.35
- 0.21→2.00+1.79
- 0.05→2.00+1.95
- 0.11→0.80+0.69
Downloads the radial above as a PNG with the edition stamped into the image.
Three rules, twenty seconds
Rank
Every robot is read against the same 40 capability checks, each scored 0–5. The Index score is the average across all of them.
Evidence
A vendor claim or a staged demo supports less than an independent report, a customer trial or a measured result. Weaker sources are capped, not ignored.
Unknown
A check with no public evidence counts as 0 in the average, and is shown as unknown everywhere else. It is a gap in the record, not a verdict on the robot.
This ranks the public record, not undisclosed internal capability. Read the full methodology for anchors, evidence ceilings, labels, levelling and the exact score calculation.
Questions about the ranking
- What is the most advanced humanoid robot?
- On the current edition of the Cyborg Index, the highest Capability score belongs to the humanoid at rank one in the table above. That is a statement about the public record, not about undisclosed internal capability: it is the humanoid with the strongest evidenced read across all forty sub-axes at the time the edition was signed off.
- How are humanoid robots ranked?
- Every humanoid is read against the same framework: eight capability axes, forty sub-axes, each scored 0–5 from public evidence. The sub-axis scores are levelled across the cohort so a 3 means the same thing on every robot, then averaged into a single Capability score. Rank is that score, descending. Full methodology.
- Why does missing evidence count as zero?
- A sub-axis with no public evidence scores 0 over a fixed denominator of forty. Skipping it instead would reward silence: a vendor that publishes nothing would be measured on a shorter, easier exam than one that publishes widely. Zero is a statement about the record, not a claim that the robot cannot do the thing.
- Are vendor demonstrations treated as proof?
- They are counted, but they are capped. Each sub-axis carries an evidence ceiling: vendor-claimed and demo material can only lift a score so far before independent reporting, named deployments or operational results are needed. Vendor material is never presented as validated evidence.
- How often is the ranking updated?
- Monthly. Each edition is dated and kept as published; scores are not silently edited after sign-off. Corrections are carried into the next edition with the source that prompted them.
- Can vendors request a correction?
- Yes. If a score misreads the public record, or a deployment, certification or field result is missing, send the evidence and it is re-read in the next edition. Corrections are made against sources, not against requests. Send a correction.
- Why are some major humanoids ranked lower?
- Visibility is not evidence. A widely covered humanoid with few named deployments, little independent testing and no published safety or field results will score below a less famous one with an operational record. The framework is identical for every entry, so the gap is in the record, not in the reading.
If a score misreads the public record, or a deployment, certification or field result is missing, send the evidence and it will be re-read in the next edition. Corrections are made against sources, not against requests.