Tesla Optimus has plenty of videos and presentations but, unlike several other humanoids, no consolidated official spec sheet. That makes it easy to confuse planned values, prototype figures and future targets. This article organizes the changes generation by generation based on IEEE Spectrum’s coverage of the launch events and Wikipedia’s timeline (including the original reports it cites), and separates what has been confirmed from what has not.
The generations at a glance
| When | Stage | What was disclosed |
|---|---|---|
| Aug 2021 | First announcement at AI Day | Planned 173 cm height, 57 kg weight, 20 kg carrying capacity |
| Sept 2022 | AI Day prototypes | Development platform and latest-generation prototype; first untethered walk on stage |
| Dec 2023 | Gen 2 video | Slimmer body, improved hands and movement; 11-DoF hands |
| June 2024 | Production remarks | Limited production in 2025; goal of 1,000+ units in Tesla facilities |
| Oct 2024 | “We, Robot” event | Reported price estimate around $30,000; teleoperation use noted |
| Nov 2024 | Hand upgrade | Third-generation hand with 22 DoF announced |
| Apr 2026 | Q1 earnings call | Gen 3 to be unveiled near production start; production-line design capacities announced |
What the 2022 prototype confirmed
According to IEEE Spectrum, three versions appeared at AI Day 2022: the 2021 concept, a development platform built with semi-off-the-shelf actuators, and a latest-generation unit with Tesla-designed actuators, battery and control system. The latest unit was not yet ready to walk, but Tesla said it showed the degrees of freedom expected in the production version.
The disclosed architecture:
- Degrees of freedom — 28 in the body, plus 11 in each hand
- Actuators — six types, three rotary and three linear; no force or torque figures were given
- Battery — 2.3 kWh, 52 V pack, described by Tesla as sized for a full day of work
- Hands — five fingers, with the fingers and thumb driven by six cable actuators per hand and springs providing the opening force; chosen for simplicity and fewer parts, and designed to carry a 9 kg bag
- Falls — the structure was designed in simulation to minimize damage when the robot falls
Mixing rotary and linear actuators reflects the different range-of-motion and force needs of each joint. The two families and how to calculate their output are covered in Rotary vs linear actuators.
The hand: from 11 to 22 degrees of freedom
According to Wikipedia’s record, the hands in the December 2023 Gen 2 video had 11 degrees of freedom, and a third-generation hand with 22 degrees of freedom was announced in November 2024. More DoF brings a hand closer to a human’s and helps fine manipulation, but it also multiplies actuators, cable or tendon routing, control complexity and possible points of failure.
Choosing cable drive for the 2022 hand can be read as a way to keep heavy drive components out of the fingers and the hand itself light. The specific actuator layout, finger force and tactile sensing of the 22-DoF hand, however, were not confirmed by official material as of the verification date. Treat articles that state these details as fact with caution and check their sources.
Design problems that grow with hand DoF
Doubling a hand’s degrees of freedom creates design problems beyond adding joints. Public material does not show how Tesla solved them, but these are the items any hand design has to address:
- Joints versus actuated joints. A 22-DoF hand may not have an independent actuator for every joint; some joints can be coupled to move together. Check which count an announcement is using.
- Cable and tendon friction and stretch. Moving the drives out of the hand keeps it light, but friction and stretch in cables passing through the wrist reduce fingertip accuracy, and path length changes as the wrist bends.
- Sensing. More DoF means more joint angles and contact states to know before the hand can be controlled well; joint encoders and fingertip tactile sensors largely decide performance.
- Training data. The more dexterous the hand, the harder it is to record human demonstrations that map onto its joints; data gloves and teleoperation rigs must match its DoF.
- Maintenance. More cables, springs and small motors make replacement intervals and repair access a large part of operating cost.
How to read the production numbers
According to the earnings-call reporting cited by Wikipedia, in April 2026 Tesla said Gen 3 would be unveiled closer to the start of production, that its first-generation line in Fremont was designed for 1 million robots per year, and that a second-generation line at Giga Texas was designed for 10 million per year in the long term. These are design capacities of production lines, not production volumes. In June 2024, limited production in 2025 and more than 1,000 units for Tesla facilities had been mentioned; actual deployment numbers need separate confirmation.
The expert view: software is the hard part
IEEE Spectrum’s Evan Ackerman judged the 2022 presentation as neither uniquely impressive nor a disappointment, noting that the hype had outpaced the results. He cautioned that demo clips may be curated best cases with unknown levels of autonomy, and argued that the hardware is not the hard part — software that makes the robot do useful work is. He also listed Tesla’s advantages: its own factories as a use case, battery and manufacturing expertise, and resources. Teleoperation was noted at the 2024 event as well.
What to check before quoting an Optimus figure
- Which generation? Say whether it is a 2021 plan, a 2022 prototype value or a later generation.
- What kind of number? Distinguish plans and targets, prototype disclosures and design capacities.
- Where from? State whether you confirmed it from Tesla directly or it reached you through press coverage.
- Like for like? When comparing with other humanoids, make sure DoF counts use the same scope, such as including or excluding the hands.
Where actuators and hands sit in a humanoid’s cost structure is discussed in Humanoid bill of materials.
Why disclosure matters for buyers and researchers
For anyone evaluating humanoids for real work, the gap between demonstration and specification is more than an academic point. A buyer planning a pilot needs sustained payload, runtime under load, maintenance intervals and safety certifications in writing; a researcher needs to know whether a capability was autonomous or teleoperated before comparing it with published results. Until a full specification is released, the safest approach is to treat Optimus figures as the event disclosures they are, label them with their date and source, and avoid mixing them with measured values from robots that publish datasheets.