Emu3
Multimodal transformer trained only with next-token prediction, from Beijing Academy of AI.
About
Emu3 is a suite of multimodal models from the Beijing Academy of Artificial Intelligence that tokenizes images, text, and video into a discrete space and trains a single transformer from scratch with next-token prediction alone — no diffusion and no compositional CLIP+LLM stack. The team reports it surpasses SDXL on image generation, LLaVA-1.6 on vision-language understanding, and can also generate video through next-token prediction.
BAAI says it open-sourced key techniques and models, with weights distributed through its Hugging Face collection. The exact license terms on the model card should be confirmed before commercial use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Paper-reported: outperforms SDXL on image generation and LLaVA-1.6 on vision-language understanding averages across 12 benchmarks (arXiv 2409.18869).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Paper-reported SOTA claims vs SDXL/LLaVA-1.6 need independent replication.; Weights distributed via BAAI HF collection; release day not pinned.. License: BAAI says key techniques and models are open-sourced; confirm exact license on the HF model card before commercial use..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · Paper-reported SOTA claims vs SDXL/LLaVA-1.6 need independent replication.
- · Weights distributed via BAAI HF collection; release day not pinned.