Back home

GO-3

x+web 0.70
Visit source

Intro

GO-3 (GO-3 Preview) is AGIBOT's next-generation embodied AI foundation model, previewed at Innovation Day 2026 on October 29, 2026. It uses a WRAM (World Reasoning Action Model) framing that combines understanding, prediction and action, with reported lead scores on RoboDojo, RoboTwin and RoboColiseum.

Official: AGIBOT article · Thread: @techniahqrobot · CN: @wangzailry

Explain

Key points

  1. Embodied foundation model: aimed at robot control/reasoning, distinct from WITA-Omni (interaction) and Link-U (OS).
  2. WRAM: company framing that folds world understanding, prediction and action into one modeling approach; demos stress zero-/few-shot behaviors.
  3. Benchmarks cited: RoboDojo, RoboTwin, RoboColiseum (company-reported leads).
  4. Open source: speakers said a fuller open release is planned by year-end; current form is a Preview.

Status

Preview / announcement stage—watch for weights, API or robot firmware packages.

Usage

What to do now

  • Follow AGIBOT channels for GO-3 Preview access and the promised open-source drop.
  • If you evaluate embodied models, note the cited suites (RoboDojo / RoboTwin / RoboColiseum) and re-run on your own hardware.