BagelVLA: Enhancing Long-Horizon Manipulation via Interleaved Vision-Language-Action Generation
Original research
MODEL PROFILE · Hybrid WAM / VLA
BagelVLA
Interleaves language planning, visual forecasting and action prediction.
Research organisationTsinghua University / ByteDance Seed
Research approachHybrid
Published year2026
BackboneBAGEL unified multimodal model
Overview
BagelVLA integrates language-level task decomposition with future visual forecasts and robot action chunks. It is a hybrid predictive-action architecture, not simply a video world model.
Source record
Core research identity and broad method were source-checked as of 2026-11-03. Independent benchmarking, commercial availability, weights and licensing have not been assessed.