Figure 03 vs Tiangong Ultra: 2026 Capabilities and Costs

Figure 03 Hardware and Helix Autonomy Stack

Figure AI released the Figure 03 in October 2025 with a redesigned chassis built around the Helix vision-language-action model. The robot carries cameras that deliver twice the frame rate of prior versions, one-quarter the latency, and a 60% wider field of view, while fingertip sensors register forces as low as three grams. These upgrades support whole-body tasks without task-specific fine-tuning. Helix 2.5, pretrained on Figure’s Index human-video dataset, completed three full tasks across 30 previously unseen Bay Area homes with a reported 56% success rate. The same pretraining step lifted overall task success from 9% to 56%. BMW has already moved Figure 02 units into an 11-month body-shop pilot at Spartanburg that supported production of more than 30,000 X3 vehicles; the newer Figure 03 is slated for a logistics sequencing role at the same plant. Company statements place BotQ factory capacity at up to 12,000 units per year, though actual customer shipments remain limited to pilot volumes.

Tiangong Ultra Mobility and Industrial Focus

Tiangong Ultra, developed by the Beijing Humanoid Robot Innovation Center, gained attention at the 2026 World Humanoid Robot Games in Beijing. The robot recorded an 8.64-second 100 m sprint in the final, improving on its own semifinal time of 8.86 seconds and beating Usain Bolt’s 9.58-second human record by nearly one second. Earlier rounds showed progressive gains from 9.39 seconds in the opening heat. Engineers used three separate simulation-trained controllers that adjusted joint torques and balance in real time, allowing the robot to maintain upright posture under high-speed loads. The same platform also posted a 38.15-second 400 m time and a 2.88 m standing high jump during the event. Post-competition statements from X-Humanoid indicate the immediate engineering priority has shifted to reliability and practical manipulation rather than further speed records. UK distributor The Robot Group opened enquiries on 19 October 2026 for both Tiangong and Tianyi models aimed at manufacturing, logistics, hospitality and infrastructure projects.

Pricing Structures and Deployment Economics

Figure 03 operates under an enterprise-only RaaS model with no public list price. Reports from the BMW pilot cite roughly $25 per robot-operating-hour. The company has not disclosed hardware purchase options or multi-year lease terms. In contrast, Chinese platforms including Tiangong Ultra target volume buyers with lower upfront hardware costs, although exact figures for the Ultra variant remain unpublished. Comparable Unitree G1 units list at $13,500–$16,000, and daily rental rates for Chinese humanoids have fallen to as low as 289 yuan (approximately $40) in some classified listings. Tiangong Ultra carries no verified customer deployments as of October 2026; all public activity consists of competition demos and distributor outreach. Figure 03 sits in paid pilots with at least one named automotive customer and approximately 1,000 units built by July 2026 according to company statements.

Task Suitability and Second-Order Implications

Figure 03’s sensor suite and Helix training pipeline favor unstructured home and light logistics environments where zero-shot generalization reduces per-site data collection. The 56% success rate across 30 homes suggests that scaling the Index dataset could further compress integration time, yet each new environment still requires safety validation and occasional human oversight. Tiangong Ultra’s sprint performance demonstrates robust high-speed balance and actuator durability, useful for material transport over long factory aisles, but the same hardware has not yet published manipulation benchmarks comparable to Helix’s household trials. Serviceability differences matter for fleet economics: Figure’s wireless charging and fingertip force sensing imply higher per-unit component costs, while Tiangong’s emphasis on first-principles gait optimization may lower long-term energy consumption at the expense of initial integration support outside China.

Market Context and Competitive Positioning

The October 2026 Agility Robotics SPAC merger with Churchill Capital XI, valued at $2.5 billion pre-money with a November 2 shareholder vote, highlights investor interest in robots with verified operating hours. Agility reports 65,000+ hours across nine customer sites. Figure 03 and Tiangong Ultra occupy different segments of that market: one leverages Western AI data advantages and automotive pilots, the other leverages domestic manufacturing scale and mobility records. Buyers evaluating multi-year total cost of ownership must weigh Figure’s $25/hour RaaS against potential Chinese hardware discounts that could reach 30–50% lower acquisition cost for fleets exceeding 100 units. Neither platform has released audited mean-time-between-failure data or spare-parts pricing, leaving operators to negotiate those terms during pilot contracts.

Limitations and Unresolved Questions

Figure 03’s home-trial success remains company-reported and has not been independently audited at scale. BMW deployment numbers beyond the initial pilot are not public. Tiangong Ultra’s sprint record occurred on a prepared track with crash mats; sustained performance on uneven factory floors or with payloads has not been demonstrated. UK enquiries opened in October 2026 will test whether conformity requirements and support infrastructure can close the gap between demo results and paid operations. Both programs continue to face the same core constraint: translating competition or pilot metrics into predictable, multi-shift revenue without extensive on-site engineering.

Fleet Planning Considerations for 2027

Operators planning 2027 deployments should model integration costs separately from hardware or RaaS fees. Figure’s Helix stack may reduce data-collection overhead in variable environments, but each additional home or line still incurs valida