Genesis AI released GENE-26.5, a foundation model for humanoid robots that autonomously cooks a tomato and egg stir-fry. The French startup's approach uses large-scale human operation data pretraining combined with simulation closed-loop evaluation.
Key facts
- GENE-26.5 autonomously cooks tomato and egg stir-fry.
- Robot cracks eggs, cuts tomatoes, makes smoothies.
- Solves Rubik's cubes and organizes cables.
- Uses human data pretraining and simulation closed-loop.
- Released by French startup Genesis AI.
Genesis AI, a French robotics startup, released its first foundation model GENE-26.5, featuring a humanoid robot that autonomously cracks eggs, cuts tomatoes, makes smoothies, solves Rubik's cubes, and organizes cables [According to Genesis AI Releases GENE-26.5]. The company's approach: large-scale human operation data pretraining combined with simulation closed-loop evaluation, moving robot manipulation toward a foundation model training paradigm.
Unique take: Unlike most humanoid robot demos that focus on locomotion or simple pick-and-place tasks (e.g., Tesla Optimus walking, Figure 01 folding clothes), GENE-26.5 targets complex multi-step manipulation with deformable objects (eggs, tomatoes) and fine motor control (Rubik's cube). This shifts the benchmark from stability to dexterity.
The robot completed the entire stir-fry cooking process without human intervention, including cracking an egg and slicing tomatoes—tasks that require precise force control and real-time adaptation. The foundation model was pretrained on a large dataset of human operation data, then fine-tuned via simulation-based closed-loop evaluation to handle edge cases [per the source].
Context: The release comes days after humanoid robots in Beijing completed a half-marathon endurance test to evaluate dynamic stability and energy management [2026-04-13]. While Beijing focused on locomotion, Genesis AI is betting on manipulation as the harder problem for real-world adoption.
Genesis AI did not disclose the size of the training dataset, the number of simulation iterations, or the compute budget for training GENE-26.5. The company also did not specify a timeline for commercial deployment or pricing.
Key Takeaways
- Genesis AI released GENE-26.5, a foundation model enabling a humanoid robot to autonomously cook stir-fry, solve Rubik's cubes, and organize cables.
- The approach uses human data pretraining and simulation closed-loop evaluation.
What to watch
Watch for Genesis AI to release a technical paper detailing the training dataset size, simulation iterations, and compute budget. Also track whether the company announces a commercial partnership with a kitchen appliance or food service company, which would signal real-world deployment plans.
