whatmodel.sh / image / Emu3
Emu3
↗ share
+ contribute
Emu3 is an open-weights image model from BAAI, released September 2024. Its published size is 8B parameters. Licence: apache-2.0.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Maker's paper (BAAI). 8B = the Emu3 decoder-only transformer (shared by Emu3-Gen/Emu3-Chat); the separate vision tokenizer is not included.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Emu3, out of 197 with a score, then more from BAAI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 162 |  JanusFlow-1.3B | 36±6 | comparison only | ||
| 163 |  PixArt-α (PixArt-XL-2-1024-MS) | 36±2 | local | ||
| 164 | Emu3 | 36±2 | local | ||
| 165 |  Ovis-U1-3B | 35±6 | comparison only | ||
| 166 |  IF-XL | 35±2 | local | ||
| more from BAAI | |||||
| 48 | Emu3.5-Image | 60±5 | local | ||
| 69 | Emu3.5 | 57±5 | local | ||
| 136 | OmniGen2 | 42±2 | api $0.15/image +local | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #103 of 126 · 98% of its score · 6 tests
Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ), Stable Diffusion XL 0.9 (base) here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| T2I-ReasonBench (overall) | 41.7% ✎ | ▅▂█▂▂▂····▂▂ beats 0/13 | 45% |
| GenExam (relaxed overall, GPT-5 judge) | 8.8% ✎ | ▄█▂▂▂▂▆▄·▂·▂ beats 1/16 | 32% |
| MMMG knowledge-image generation (MMMG-Score, average over education levels) | 6.1% ✎ | ▃··▃·█▃▃▅··▃ beats 0/9 | 11% |
| WISE (WiScore overall, GPT-4o judge) | 39.0% ✎ | ▄·▆▂▆█▄▆▂▄·▄ beats 6/22 | 10% |
| GenEval lab | 54.0% ✎ | ▇▃▅▇▁▄▅▃█▃▃▅ beats 8/39 | — |
| DPG-Bench lab | 80.6% ✎ | ▄·▂·▂▆█▄▂▄▂▂ beats 1/17 | — |
$ where to run it
Other ways to run it
- Run it as published · needs ~22.2 GB, NVIDIA only · from github.com/baaivision
git clone https://github.com/baaivision/Emu3 && cd Emu3 && pip install -r requirements.txt && python image_generation.py
Official weights and documentation: huggingface.co/BAAI ↗
$ how the score was worked out · T2I-ReasonBench: Benchmarking… 45%, GenExam: A Multidisciplinary… 32%, MMMG: A Massive, Multidiscipl… 11% and 1 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026) | 1 | 45% |
 GenExam: A Multidisciplinary Text-to-Image Exam (arXiv v5) | 1 | 32% |
| MMMG: A Massive, Multidisciplinary, Multi-Tier Generation Benchmark for Text-to-Image Reasoning (arXiv:2506.10963v2) | 1 | 11% |
| WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4) | 1 | 10% |
| prior · pull toward the middle | 2% | |
| weighted average | = 35.6 |
not counted (2)
| BAAI — Emu3 technical report | DPG-Bench [correct] | 80.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| BAAI — Emu3 technical report | GenEval [correct] | 54.0% | fewer than three models have this test yet, or this source ran it for this model alone |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json