Genesis AI Goes Full Stack with Human-Shaped Hands

Genesis AI just unveiled its first model, GENE-26.5, paired with a set of robotic hands the company designed in-house. According to TechCrunch AI, the Khosla-backed startup, which raised a $105 million seed round in July 2025, decided to build the entire stack rather than just the AI brain. Co-founder and CEO Zhou Xian told TechCrunch the team realized a better model needed control over the hardware too, so they went full stack.

What stands out here is the bet on hand design. Most robotics companies stick with two-finger grippers. Genesis built a hand that matches the size and shape of a human hand, which co-founder Théophile Gervet says shrinks the gap between training data and real-world conditions.

What Genesis Launched

  1. GENE-26.5 foundation model. Named for May 2026, with more iterations planned. The team built a simulation system to speed up evaluation, which Zhou calls the real bottleneck for model iteration.
  2. In-house robotic hand. Human-sized, human-shaped, designed to close the embodiment gap that has slowed robotics research. Gervet, a former Mistral AI research scientist, says the form factor unlocks much more training data.
  3. A sensor-loaded data collection glove. It mirrors the robotic hand and is light enough to wear during a normal workday, similar to security gloves already used in labs and manufacturing.
  4. Demo task library. The video shows the system cracking eggs, slicing tomatoes, blending smoothies, playing piano, solving a Rubik’s cube, and running lab work. Gervet’s favorite is cooking because it chains many hard subtasks together.

Why the Glove Matters

Genesis is pitching the glove as a way for customers to collect training data while employees do their actual jobs. Gervet said the company is in talks with potential customers across pharma labs and manufacturing.

There’s an obvious tension. Workers wearing data-collection gloves are training the systems that could eventually replace them. Asked whether those workers would get extra pay, Gervet said that will be between Genesis customers and their employees. “We haven’t nailed the details yet,” he told TechCrunch.

If customers refuse to share the captured data, Genesis has backup plans. The model already trains on what the company calls “massive amounts of human-based internet videos,” and Genesis can pay third-party partners to build out its human skill library.

How It Stacks Up

Genesis isn’t alone in this space. TechCrunch AI notes that Physical Intelligence and Skild AI work on similar problems at the AI-robotics intersection, and Zhou himself acknowledged there are probably 50 to 100 robotic hand companies out there. The wager is that owning both the model and a human-shaped hand produces compounding advantages competitors can’t easily match.

Google’s former CEO Eric Schmidt, who backed the round, called the launch “an important milestone for their team and the robotics industry more broadly.” Other backers include Eclipse, Khosla Ventures, Bpifrance, HSG, Xavier Niel, Daniela Rus, and Vladlen Koltun.

Where Things Stand

Genesis now runs offices in Paris, California, and London with 60 employees split roughly 40-45% in Europe and 50-55% in the U.S. The team is hiring across all three locations. Gervet pointed to European talent density as a deliberate strategic choice.

The near-term roadmap includes a first general-purpose robot, which Zhou said will be a full-body system rather than just hands. He insisted the broader goal hasn’t shifted: build the most capable robotic system possible.

What Comes Next

The simulation pipeline plus the human-shaped hand plus a wearable data collection device is a coherent story. If Genesis can convince enterprise customers to deploy the glove at scale, the data flywheel could accelerate fast. The open questions are whether the labor side of the equation gets sorted, whether the full-body robot delivers on the demo’s promise, and whether human-shaped hardware actually pays off versus simpler grippers in commercial settings.

More details on the launch and the demo video are available at the original TechCrunch AI report.

Scroll to Top