Khosla-backed Genesis AI Unveils Full-Stack Robotics Solution

Genesis AI, a startup that secured a $105 million seed round to develop foundational AI for robotics, has launched its inaugural model, GENE-26.5, featuring unexpectedly dexterous hands. In a demonstration video, the company highlighted various advanced tasks executed by its in-house-designed robotic hands.

“The model has always been the primary objective, as superior intelligence stems from better models,” Genesis co-founder and CEO Zhou Xian told TechCrunch. However, the company soon recognized the necessity of hardware control. “Consequently, we chose to pursue a full-stack approach,” he explained.

Other well-funded firms operate at the intersection of AI and robotics, including Physical Intelligence and Skild AI. Xian acknowledged that “there are likely 50 to 100 robotic hand companies in existence.” Nevertheless, he and co-founder Théophile Gervet believe that building their own hardware will provide a competitive advantage.

The critical distinction is that Genesis’ hand matches the size and shape of a human hand, unlike the two-finger grippers commonly used by other robotics companies, thereby narrowing the gap with real-world scenarios.

“This enables us to gather significantly more data than previously possible, allowing us to train models capable of executing a wider array of tasks,” said Gervet, a former research scientist at Mistral AI who now serves as Genesis’ president.

Among the physical manipulation tasks demonstrated in the video below, Gervet’s personal favorite is cooking, as it demonstrates the robot’s ability to complete a complex sequence of difficult actions, such as cracking an egg and slicing a tomato. Genesis has also tasked its robots with preparing smoothies, playing the piano, and solving a Rubik’s cube—a common robotics benchmark.

Other tasks, such as laboratory work, represent more immediate commercial applications for Genesis’ technology. However, the underlying process is equally crucial: the startup has also developed a sensor-equipped glove that serves as a real-life counterpart to its robotic hand, collecting data that is easier to utilize.

“Our concept was that by designing a robotic hand that closely mimics a human hand, we could instantly unlock vast amounts of human data without worrying about the so-called ‘embodiment gap’ in robotics research,” Xian stated.

Others have attempted to address this challenge; the key innovation lies in how Genesis integrates this with its model. The current version, named GENE-26.5 for May 2026, is expected to undergo numerous iterations thanks to the simulation system they have developed. “The primary bottleneck for model iteration speed is evaluation. This significantly accelerates our model training process,” he noted.

Beyond simulation, data remains essential for training models that enable robots to perform more tasks. This is where Genesis’ glove proves valuable. Gervet explained that, unlike cumbersome data collection devices, it is lightweight, easy to wear, and relatively inexpensive to produce, similar to security gloves already used in various industries.

“We are currently in discussions with numerous customers, and a key value proposition of the glove is that it allows users to wear the data collection device while performing their daily jobs, whether as a pharmaceutical lab technician or in manufacturing,” Gervet said. This approach is further supported by ‘egocentric video data,’ where individuals film themselves performing tasks.

Nevertheless, it remains uncertain whether workers would be willing to wear the gloves and cameras that could train robots to replace them, or if they would receive additional compensation for this training. Gervet suggested that this is a matter between Genesis’ customers and their employees. “We have not yet finalized these details,” he remarked.

Regardless, workers may choose not to share this data with the startup, a possibility the founders acknowledge. However, Genesis has other avenues to build its ‘human skill library,’ including paying third-party partners to collect data. According to a press release that did not mention compensation, its model is already trained on “massive amounts of human-based internet videos.”

Combined with its simulation system, this could help Genesis reduce the costs of its technology for real-world applications, such as those demonstrated. “This represents a significant milestone for their team and the broader robotics industry,” said Eric Schmidt, Google’s former CEO and an investor in the startup.

In July 2025, just months after its founding, the startup emerged from stealth with a $105 million seed round co-led by Eclipse and Khosla Ventures. Additional backers included Bpifrance, HSG, and individuals such as Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun.

This funding enabled Genesis to expand its workforce. With offices in Paris and California, the company has also opened a location in London. “A major reason for establishing a presence in Europe is the high talent density across the continent,” Gervet stated. The team of 60 people is distributed with “40-45% in Europe and 50-55% in the U.S.,” and the startup is currently hiring in all three locations.

In addition to hiring, the company plans to unveil its first general-purpose robot soon, which Xian told TechCrunch will be a full-body robot, not just hands. However, he emphasized that the roadmap remains unchanged.

“Our objective is to create the most capable robotic system,” he concluded.