ByteDance Unveils 3B Model Lance
·2026.05.19 21:05
Key point
ByteDance has released Lance, a 3B open-source model that combines image and video understanding, generation, and editing.
1 / 2
Details
Lance is a 3B native unified multimodal model released by ByteDance.
It is designed to handle understanding, generation, and editing of images and videos within a single framework, and was trained from scratch using a staged multi-task recipe.
The official page and model card also disclosed the following:
- Training budget: 128 GPUs or fewer
- License: Apache-2.0
- Benchmarks:
GenEVAL,DPG-Bench,GEdit-Bench,VBench,MVBench
Key results are as follows:
- Tied for the highest overall score on
GenEVALamong compared unified models - Highest average score on
GEdit-Benchamong compared unified models - Highest average score on
MVBenchamong compared unified models
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.