whatmodel.sh / image / BLIP3-o 8B

BLIP3-o 8B

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

BLIP3-o 8B is an open-weights image model from Salesforce Research, released May 2025. Its published size is 8B parameters. Licence: apache-2.0.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

Maker's rounded size label for the whole BLIP3-o 8B model: a frozen Qwen2.5-VL-7B-Instruct backbone plus diffusion transformers totalling 1.4B trainable parameters (no exact total is published). The released weights are the 8B model retrained on open-source data only.

38.4±2.8#111 of 144 image modelsfrom 3 benchmark results across 3 sources

Score within the LMArena text-to-image comparison set only.

$ how it compares

The image models ranked either side of BLIP3-o 8B, out of 197 with a score, then more from Salesforce Research.

#modelscorewhereRAMreleased
149Stable Diffusion 3 Medium Stability AI40±2api $0.035/image +local~11GB2024-06
150Kolors Kuaishou (Kolors team)39±3local~22GB2024-07
151BLIP3-o 8B Salesforce Research38±3local~41GB2025-05
152Stable Cascade Stability AI38±4local~13GB2024-02
153Janus-Pro-7B DeepSeek38±2api $0.0020/image +local~18GB2025-01
more from Salesforce Research
132BLIP3o-NEXT-GRPO-Text-3B Salesforce Research42±3comparison only

see all image models · find the ones that fit your machine

$ what it's good at

text to image▂▅▅▃▄▄█▃▁▂▁▁ #95 of 126 · 97% of its score · 3 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. Trails SDXL-Lightning (4-step UNet), Stable Cascade here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
GenExam (relaxed overall, GPT-5 judge)6.7% ✎▄█▂▂▂▂▆▄·▂·▂ beats 0/16hard · avg 26%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 64%sharp
WISE (WiScore overall, GPT-4o judge)52.0% ✎▄·▆▂▆█▄▆▂▄·▄ beats 15/22medium · avg 60%▇▇▇▇▇▇▇ 17%at its level
XTC-Bench generation (Overall Gen., scene-graph node agreement)54.5% ✎▄█·▄·······▄ beats 3/4easy · avg 70%▇▇▇▇▇▇ 16%at its level

$ where to run it

Other ways to run it

  1. Run it as published · needs ~41.3 GB, NVIDIA only · from github.com/JiuhaiChen
    git clone -b main https://github.com/JiuhaiChen/BLIP3o.git && cd BLIP3o && conda create -n blip3o python=3.11 -y && conda activate blip3o && pip install --upgrade pip setuptools && pip install -r requirements.txt && python inference.py $(python -c "from huggingface_hub import snapshot_download; print(snapshot_download(repo_id='BLIP3o/BLIP3o-Model-8B', repo_type='model'))")

Weights and documentation: huggingface.co/BLIP3o ↗

$ how the score was worked out · GenExam: A Multidisciplinary… 64%, WISE: A World Knowledge-Infor… 17%, Beyond Accuracy: Benchmarking… 16%

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

sourceresultsshare of the scorelicence
GenExam: A Multidisciplinary Text-to-Image Exam (arXiv v5)1▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 64%CC-BY-4.0
WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)1▇▇▇▇▇ 17%CC-BY-SA-4.0
Beyond Accuracy: Benchmarking Cross-Task Consistency in Unified Multimodal Models (XTC-Bench, arXiv:2604.25072v1)1▇▇▇▇▇ 16%CC-BY-4.0
prior · pull toward the middle▇ 3%
weighted average= 38.4

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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