Blog or model card only

GO-3 Preview is AGIBOT’s embodied foundation model unveiled on October 29, 2026. Its WRAM architecture unifies understanding, prediction and action for robot task generalization and manipulation. Full open sourcing is planned for the end of 2026.

Research material status

Blog or model card only

Official introductory materials are recorded; no version-specific paper or technical report has been verified.

Last checked: 2026-11-05

Papers & reports

No published papers recorded.

Access and availability

Release information

SourceVerified on · 2026-11-05

At a glance

Model type
Robot foundation model
Released
October 29, 2026

Model card

GO-3 Preview is AGIBOT’s embodied foundation model unveiled on October 29, 2026. Its WRAM architecture unifies understanding, prediction and action for robot task generalization and manipulation. Full open sourcing is planned for the end of 2026. AGIBOT reports first place on RoboDojo, RoboTwin and RoboColiseum. Original scores and test conditions remain unverified. Parameter count and actual weight/code release dates are unconfirmed. Planned open sourcing does not establish current access. [AGIBOT](https://www.agibot.com/article/231/detail/131.html) · [Jiemian](https://www.jiemian.com/article/15262681.html) AGIBOT’s October 29 announcement describes WRAM as unifying task understanding, future-state prediction and action generation. Genie Evolver combines 100 physical robots, physics simulation and world models for hybrid post-training; Genie Master is the deployment-stage embodied agent framework. Benchmark leadership and full open sourcing by end-2026 are company claims. License, release scope, evaluation protocols and baselines, reproducible real-robot success rates and specific Genie/YuanZheng hardware compatibility remain unverified. The November 3 PRNewswire release repeats the company announcement, not an independent evaluation. [AGIBOT](https://www.agibot.com/article/231/detail/133.html) · [PRNewswire](https://www.prnewswire.com/apac/news-releases/agibot-unveils-go-3-preview-takes-top-spot-on-three-embodied-ai-benchmarks-302927277.html)

Before you use this version: AGIBOT’s October 29 announcement describes WRAM as unifying task understanding, future-state prediction and action generation. Genie Evolver combines 100 physical robots, physics simulation and world models for hybrid post-training; Genie Master is the deployment-stage embodied agent framework. Benchmark leadership and full open sourcing by end-2026 are company claims. License, release scope, evaluation protocols and baselines, reproducible real-robot success rates and specific Genie/YuanZheng hardware compatibility remain unverified. The November 3 PRNewswire release repeats the company announcement, not an independent evaluation. Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.

Submit a source or correction