Generalist Ai · v1.5 · depuis 19.-20. August 2026 · 5× · vu le 24 août 2026
GEN-1.5 is a robot foundation model from the US startup Generalist AI (founded in 2024 by former Google DeepMind and Boston Dynamics researchers), unveiled on August 19/20, 2026. The multimodal model processes video, sensor, language and proprioceptive inputs and produces 100 Hz action trajectories; it can learn new physical tasks from a single 3-12 second demonstration ("physical prompting") without gradient updates or fine-tuning. Across ten manipulation tasks it averaged a 59% one-shot success rate, rising to 83% with additional training (10 steps, about five minutes of data). No public pricing model, license, or self-service deployment option is documented so far; access is currently only available via early-access requests to the company.
Fonctionnalités
| Deployment Model | No self-service deployment; access via early-access request by email to the company, live demos at trade shows (e.g. Automate 2026) |
| Use Case Scope | Physical manipulation tasks for robots (e.g. grasping, tool use, household/industrial tasks) via one-/few-shot learning from demonstration |
| Platform | Robot foundation model (multimodal: video, sensor, language, proprioception), 100 Hz action output, 30-second context window |
| Price | No public pricing disclosed; access currently by early-access request only |
| Release Date | August 19, 2026 (published) |