88Z AI, 2 November — Agibot has released the GO-3 Preview embodied intelligence model, achieving scalable learning across pre-training, post-training, and deployment. The model secured first place in RoboDojo, RoboTwin, and RoboColiseum benchmarks. It uses WRAM to absorb human first-person data during pre-training, employs Genie Evolver for mixed training with real robots, simulation, and world models, and leverages Genie Master to accumulate experience in real-world settings like convenience stores.
At a glance
- GO-3 Preview ranked first in RoboDojo, RoboTwin, and RoboColiseum benchmarks
- Achieves full-chain scalable learning and experience accumulation via WRAM, Genie Evolver, and Genie Master
The update
Agibot has officially launched the GO-3 Preview embodied intelligence model, which integrates scalable learning across pre-training, post-training, and post-deployment stages. In terms of performance, GO-3 Preview achieved first place in three authoritative benchmarks: RoboDojo, RoboTwin, and RoboColiseum. Technically, the pre-training phase utilizes WRAM technology to absorb approximately ninety percent of human first-person perspective data. The post-training phase employs the Genie Evolver system, combining real robot operations, simulation environments, and world models for mixed training. After deployment, the model uses Genie Master to continuously accumulate success and failure experiences in real-life scenarios such as convenience stores, enabling continuous evolution and capability optimization.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 机器之心Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.