Tesla Optimus vs Figure 02: A 2026 Head-to-Head on Data, Not Hype

Figure AI's Figure 02 finished an 11-month BMW Spartanburg deployment in November 2025, contributing to 30,000 X3 vehicles. Tesla's Optimus Gen 3 has roughly 300 units in what Elon Musk himself called a "learning phase" on the Q4 2025 earnings call. These two robots are the most-watched US humanoid programs, and the gap between them is no longer a question of marketing — it is a question of what each one has actually done on the floor. See also the Figure 02 vs Optimus Gen 3 spec sheet and the consumer-focused Figure 03 at the $20,000 price point.

This is a head-to-head on the data we can verify. Specs, deployments, software stacks, costs, and production capacity. Where a number is a manufacturer claim that has not been independently checked, the text says so.

Spec sheet, side by side

Both robots are bipedal humanoids in the 168 cm tall, ~57–70 kg weight class, with 28 degrees of freedom in the body. From there the design choices diverge.

| Metric | Optimus Gen 2 | Figure 02 | |---|---|---| | Height | 168 cm | 168 cm | | Weight | 57 kg | 70 kg | | Body DoF | 28 | 28 | | Hand DoF | 11 (revised to 22 in 2024) | not publicly specified per hand | | Rated payload | 20 kg | 25 kg | | Battery | 2,300 Wh, 480 min runtime (manufacturer) | 2,250 Wh, 300 min runtime (manufacturer) | | Top walking speed | 2.2 m/s | 1.2 m/s | | Compute | Tesla FSD-derived custom SoC | Dual NVIDIA RTX GPU modules | | Vision | 8 cameras, Autopilot-heritage | 6 RGB cameras | | LiDAR | No | No | | ROS support | None | None | | Status | Internal fleet only | Customer deployments + preorders |

The 13 kg weight difference is partly compute hardware: dual NVIDIA RTX modules in Figure 02 are heavier than Tesla's purpose-built SoC. Figure trades weight for headroom on the neural-network side. Tesla trades the other way and stays leaner.

Figure 02 rates a higher payload (25 vs 20 kg). Tesla rates a longer runtime (480 vs 300 minutes), but neither company has published a duty-cycle table that links payload to runtime, so the two runtime figures are not directly comparable.

Top walking speed favors Tesla on paper at 2.2 m/s versus Figure's 1.2 m/s. In practice the BMW Spartanburg cycle was paced by an 84-second cycle time on the assembly station, not by walking speed.

Deployment evidence: what actually happened

BMW Spartanburg, January–November 2025. Figure 02 ran 10-hour shifts five days a week. Operating hours: 1,250. Parts handled: more than 90,000. Vehicles X3 produced with the robot's contribution: 30,000+. Placement accuracy reported by Figure and BMW: above 99 percent per shift. Steps logged: roughly 1.2 million. BMW Group's official press release on November 25, 2025 ended the trial as a success and announced the next deployment at Plant Leipzig.

This is the first general-purpose humanoid program with publicly documented production-line work at automotive scale. Whether the next phase is paid commercial rollout or a longer trial has not been fully disclosed.

Tesla Fremont and Giga Texas, mid-2024 to present. Tasks shown include battery cell sorting from the 4680 line, parts handling on the Cybertruck floor, and quality inspection of structural castings. By Tesla's own Q4 2025 earnings commentary (January 28, 2026), the deployed fleet is around 300 Gen 3 units across both factories, and the explicit purpose is to generate teleoperation traces and autonomous failure logs for the next training cycle. There are no published numbers for cycle time, parts moved, or shift hours.

The contrast is structural. Figure built a single deep deployment with a customer who agreed to publish results. Tesla built a wider internal fleet without a customer and without published metrics. Each company will defend its choice. Investors and integrators have to weigh them differently.

Software: closed black box versus Helix

Tesla's stack is closed. There is no ROS 2 support, no SDK for third parties, no Gym-style benchmark environment. The compute platform is a custom SoC derived from FSD silicon. The training data pipeline is internal. The only way to extend Optimus behavior is to be Tesla.

Figure has taken a different bet on visibility. The Helix model is a vision-language-action system that learns from demonstration video. Figure has shown towel-folding learned from 80 hours of footage. In October 2025 Figure announced Helix 02, the upgraded version, alongside Figure 03 — a smaller, $20,000 home-focused robot that ships with the same software stack as the enterprise platform. Figure publishes engineering blog posts on the Helix architecture in enough detail to be discussed in IEEE papers and academic VLA roundups.

That does not make Figure open in the ROS or open-source sense — there is no SDK and no commercial customer integration outside Figure's direct supervision. But the externally visible technical story is denser than Tesla's, which gives independent observers more to verify and benchmark against.

For a researcher choosing a platform today, neither one is the right answer. Both stacks are closed at the integrator level. The practical alternative is Unitree H1 or Boston Dynamics Atlas Research, both of which ship with ROS 2 and a documented developer story.

Price and total cost of ownership

Figure 02 is a $130,000 to $150,000 enterprise platform per industry reporting. Tesla has not set a Gen 2 or Gen 3 enterprise price. The widely cited $20,000 figure is a long-term consumer target stated by Elon Musk, not a current price for any Optimus generation.

The interesting overlap is Figure 03 at $20,000, announced October 9, 2025, targeting late-2026 limited home availability. This makes the $20,000 price point real, in a Figure-branded chassis, before any Optimus consumer unit has been delivered. Tesla still has the larger projected production capacity, but Figure has the price-validated SKU first.

Total cost of ownership at production scale is harder to assess because neither company has published failure rate, mean time between failures, or expected service life. BMW Spartanburg's 11 months at 1,250 hours gives the field its first real reliability data point, and the number that matters most — how many faults happened, and how many were addressed by remote ops versus on-site service — has not been published. Until it is, TCO claims by either manufacturer remain marketing positions.

Manufacturing capacity and supply

Tesla's stated production targets through 2026 are aggressive on paper. Musk's CES 2025 projection was 50,000–100,000 Optimus units produced by year-end 2026, with Gen 3 mass production starting January 21, 2026 at Fremont. As of the Q4 2025 earnings call only about 300 had actually been built, and Musk described that fleet as a learning phase rather than production-ready output. Whether the Fremont line can scale from 300 to 50,000 in eleven months remains the most important open question for the program.

A dedicated Giga Texas Optimus factory has been announced with a stated long-range capacity of 10 million units per year and a first-line target of 2027.

Figure's BotQ factory in San Jose, opened in 2025, has a published capacity of 12,000 units per year in early 2026. That is two orders of magnitude smaller than Tesla's 2026 target, but two orders of magnitude larger than any number Figure has actually shipped. Figure has set a four-year target of 100,000 units across the BotQ network.

Two different bets. Tesla is building vertical capacity ahead of demand and counting on AI-driven cost reduction. Figure is building output to match the demand it can actually demonstrate, anchored to the BMW reference deployment.

What this comparison does not settle

Three questions remain open and matter more than any spec gap.

First, customer count outside the reference accounts. Figure has BMW publicly and a smaller set of named partners (Brookfield, others). Tesla has zero external customers — its entire fleet is captive. Neither company has published the number of paying enterprise users buying robots versus running pilots, and that number drives whether the program survives the next fundraising round.

Second, the reliability data. BMW Spartanburg's 1,250 operating hours is the first data point of its kind. The field needs at least two more reference deployments at this duration to know whether 99% placement accuracy generalizes, what the failure modes look like at scale, and whether the cost per shift competes with conventional automation.

Third, the consumer humanoid question. Figure 03 at $20,000 will either land in a small number of homes by late 2026 and validate the price point, or it will slip into 2027 and join Tesla's consumer target in the "promised but not delivered" category. If Figure ships first, that resets the conversation. If neither does, the humanoid robot story stays an enterprise story for another year.

Bottom line for buyers and researchers

If you need a humanoid in your facility in the next 18 months, Figure 02 is the only platform with documented production-line work. Tesla is not selling Optimus to external customers, and Boston Dynamics Atlas Electric is on a separate commercial timeline with Hyundai.

If you are choosing a research platform, neither one is appropriate. Unitree H1 ships with ROS 2 at around $90,000. Boston Dynamics has a research-tier Atlas program. Optimus and Figure 02 are not for sale to academia.

If you are an investor reading press releases, the numbers that matter are: customer count, paid pilots converted to production rollout, and reliability hours. Specs and price tags are downstream of those three.