AI models
Emu3.5
Training compute
9.9×10²⁴ FLOP
Parameters
34.1B
Published
Oct 30, 2025
Emu3.5 is an AI model developed by Beijing Academy of Artificial Intelligence / BAAI (China), first published in October 2025. It works in the video, multimodal, image generation and 3 more domain, on tasks such as text-to-video, image-to-video, image generation and 6 more.
Training it took an estimated 9.9×10²⁴ FLOP of compute (estimation method: operation counting). The model has 34,100,000,000 parameters. It was trained on roughly 13.6T datapoints.
Access: Open weights (unrestricted). Its weights are openly available. It is built on top of Whisper v2,Qwen3-32B. Epoch AI rates the confidence of this record as confident.
Full record
- Organization
- Beijing Academy of Artificial Intelligence / BAAI
- Country of organization
- China
- Domain
- Video, Multimodal, Image generation, Vision, Language, Speech
- Task
- Text-to-video, Image-to-video, Image generation, Text-to-image, Visual question answering, Language modeling/generation, Question answering, Speech recognition (ASR), Video description
- Training compute
- 9.9×10²⁴ FLOP
- Compute estimation method
- Operation counting
- Parameters
- 34,100,000,000
- Dataset size
- 13.6T
- Model accessibility
- Open weights (unrestricted)
- Open weights
- Yes
- Base model
- Whisper v2, Qwen3-32B
- Epoch confidence
- Confident
More from Beijing Academy of Artificial Intelligence / BAAI
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
← All ai models