BAGEL Models by ByteDance
1 model2025Up to 33k ctx
Last refreshed 2026-06-21. Next refresh: weekly.
Details
ResearcherByteDance
LicenseApache 2.0OSI-approved
Commercial useCommercial use: permitted
Models1
Released2025
Max context33k
Capabilities
VisionAll models
MultimodalAll models
About
BAGEL (Big Advanced Generalized Embodied Learner) is ByteDance Seed's open-source unified multimodal foundation model built on Qwen2.5-7B-Instruct with a Mixture-of-Transformer-Experts (MoT) architecture. It supports text understanding, visual reasoning, text-to-image generation, and image editing, trained on trillions of interleaved multimodal tokens spanning language, image, video, and web data.
Decision facts
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
1 in view
BAGEL 7BCurrent
Use when the workload needs 33k context, 7B parameters, and multimodal inputs.
2025-0533k context7B parametersmultimodal inputs
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| BAGEL 7B | Use when the workload needs 33k context, 7B parameters, and multimodal inputs. | 2025-05 | 33k context7B parametersmultimodal inputs | Current |
Release Timeline
1 release group2025-05
1 current
BAGEL 7B
Current33k context7B parametersmultimodal inputs
Specifications(1 models)
| Model | Released | Context | Parameters | Vision | Multimodal |
|---|---|---|---|---|---|
| BAGEL 7B | 2025-05 | 33k | 7B | Yes | Yes |
