product generation · 2026-07-09
NEO hands · July 2026 specification ↗
Vendor hand design and production claims scoped to the July 2026 announcement; no universal task capability is inferred.
- Hand and wrist dof
- 25
- Hand actuated dof
- 22
- Wrist dof
- 3
The evidence ledger
Model & product versions
Exact release and paper-variant identities. An unspecified checkpoint remains unknown; a newer generation never inherits an older one’s results.
model version · 2026-01-27
Helix 02 — January 2026 release ↗
Vendor architecture disclosure for the full-body VLA release; demonstrations are not independent benchmark results.
- Version
- Helix 02
- Architecture
- System 2 semantic reasoning · System 1 visuomotor policy · System 0 whole-body controller
- S1 frequency
- 200 Hz
product generation · 2025-10-09
Figure 03 — October 2025 generation ↗
Launch-specific hardware changes from Figure 02, recorded as vendor specifications with exact generation boundaries.
- Generation
- Figure 03
- Predecessor
- Figure 02
- Camera frame rate multiplier
- 2
model version
3D Diffuser Actor ↗
Paper-defined evaluation variant of 3D Diffuser Actor; tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
3D Diffuser Actor (from EL3DD) ↗
Paper-defined evaluation variant of 3D Diffuser Actor (from EL3DD); tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
3D Diffuser Actor (from EquAct) ↗
Paper-defined evaluation variant of 3D Diffuser Actor (from EquAct); tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
3D Diffuser Actor + GraspCorrect ↗
Paper-defined evaluation variant of 3D Diffuser Actor + GraspCorrect; tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
3D Foresight ↗
Paper-defined evaluation variant of 3D Foresight; tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
3D-MVP ↗
Paper-defined evaluation variant of 3D-MVP; tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
3DThinkVLA ↗
Paper-defined evaluation variant of 3DThinkVLA; tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
ABot-M0 ↗
Paper-defined evaluation variant of ABot-M0; tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned
model version
ABot-M0.5 ↗
Paper-defined evaluation variant of ABot-M0.5; tied to its measuring paper and training regime, with checkpoint identity explicitly undisclosed.
- Upstream training regime
- finetuned