Full Stack Robotics: Genesis AI Debuts GENE-26.5 Model and Custom Hand

Genesis AI, the well-funded robotics startup that emerged from stealth last July with a $105 million seed round led by Eclipse and Khosla Ventures, has unveiled its first foundation model, GENE-26.5. But the real surprise is the hardware: a five-fingered robotic hand designed in-house.

CEO Zhou Xian told TechCrunch that the model was always the goal, but the company realized it needed control over the hardware to collect the right data. "So we decided to go full stack," he said.

Why a Humanoid Hand?

Most robotics companies use two-finger grippers. Genesis AI's hand matches the size and shape of a human hand, reducing the "embodiment gap" between human and robot. This lets the team collect more data from human demonstrations and train models that can perform a wider range of tasks.

Co-founder and president Théophile Gervet, a former research scientist at Mistral AI, explained: "That lets us collect a lot more data than was previously possible, to train a model that can do many more tasks."

The Demo: Cooking, Piano, and Lab Work

In a demo video, Genesis showed its robotic hands performing complex manipulation tasks: cracking an egg, slicing a tomato, preparing a smoothie, playing the piano, and solving a Rubik's cube. Gervet's favorite is cooking because it requires a long sequence of difficult actions.

Other tasks, like lab work, point toward commercial applications. But the real innovation is behind the scenes: a sensor-loaded glove that acts as a real-time double of the robotic hand. Workers wear the glove while performing their jobs, and the data can be used to train the robot. The glove is lightweight, easy to wear, and relatively cheap to produce.

Data Collection Strategy

Genesis is in talks with customers about using the glove for data collection in industries like pharma and manufacturing. The company also plans to collect "egocentric video data" — people filming themselves doing tasks. Combined with its simulation system, this should accelerate model training.

"The real bottleneck for the iteration speed of the model is evaluation," Xian said. "So this helps us speed up model training a lot."

The Human Factor

Will workers be happy to wear the gloves and cameras that could train robots to replace them? Gervet acknowledged that compensation details haven't been nailed down. "We're in talks with a lot of customers right now," he said, but it will be up to customers and their employees to decide.

Genesis also has other avenues: paying third-party partners to collect data, and training on "massive amounts of human-based internet videos" (per a press release that didn't mention compensation).

What's Next

The company has expanded to London, with offices in Paris and California. Its team of 60 is split roughly 40-45% in Europe and 55-60% in the U.S. Genesis plans to reveal its first general-purpose robot — a full-body robot, not just hands — shortly.

"Our goal is to build the most capable robotic system," Xian said.