whatmodel.sh / image / LongCat-Image-Edit
LongCat-Image-Edit
↗ share
+ contribute
LongCat-Image-Edit is an open-weights image model from Meituan, released December 2025. Its published size is 6.27B parameters. Licence: apache-2.0.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Diffusion transformer only: OUR COUNT from the safetensors header of the official transformer weights (pinned revision; 6,270,668,864 BF16 parameters). The HF API gives no safetensors total for this repo. The 5-shard text encoder and the VAE (83.8M) are not included. The card states no parameter count. Observed 2026-10-07.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of LongCat-Image-Edit, out of 197 with a score, then more from Meituan.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 70 |  JoyAI-Image-Edit | 57±3 | comparison only | ||
| 71 |  MAI-Image-1 | 57±3 | comparison only | ||
| 72 | LongCat-Image-Edit | 56±2 | api $0.15/image +local | ||
| 73 |  Seedream 3.0 | 56±2 | comparison only | ||
| 74 |  P-Image-Edit | 56±4 | api $0.010/image | ||
| more from Meituan | |||||
| 88 | LongCat-Image | 53±4 | api $0.13/image +local | ||
$ what it's good at
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #35 of 62 · 73% of its score · 6 tests
Beats FLUX.2 [flex], Imagen 4, GPT Image 1 Mini here, though they rank higher overall. Trails FLUX.2 [klein] 9B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Edit-Compass (EN, overall) | 53.0% ✎ | ▃▃▂·▃▆▂█·▃·▃ beats 14/20 | 39% |
| CEdit-Bench-CN overall (G_O, GPT-4o judge) lab | 76.5% ✎ | ▃·········█▂ beats 6/7 | 16% |
| CEdit-Bench-EN overall (G_O, GPT-4o judge) lab | 76.7% ✎ | ▄·▄···▄██··▄ beats 6/7 | 10% |
| GEditBench v2 (overall Elo) | 1001.0 ✎ | ▃▂····▂█▂▃▂▃ beats 5/14 | 7.8% |
| GEdit-Bench-EN overall (G_O, GPT-4.1 judge) lab | 76.4% ✎ | ▄·····█·▄▄·▄ beats 5/5 | — |
| ImgEdit-Bench overall (GPT-4.1 judge) lab | 90.0% ✎ | no other model yet | — |
other tests3 not in a task · 25% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| ChartSync chart editing (Overall) | 47.3% ✎ | ▂··▂█▂▄▂▂·▂▂ beats 2/12 | 18% |
| PaintBench precise editing (mIoU, 20-task average) | 3.6% ✎ | ▃·····█▅▃·▃▅ beats 2/9 | 4.2% |
| ViGoR-Bench zero-shot visual reasoning, image models (Result Avg) lab | 32.2% ✎ | ▂···▂·▂█▆▅·▂ beats 8/15 | 3.1% |
$ where to run it
Through an API
api · fal.ai · $0.15/image · fal.ai ↗ · price sourceUSD for one 1-megapixel output: fal bills this endpoint $0.15 per megapixel (price = rate × 1 MP; fal's rounding for 1024×1024 ≈ 1.05 MP is not stated; larger outputs cost proportionally more). Product: fal.ai serverless image editing. API model: fal-ai/longcat-image/edit (LongCat Image Edit). Settings: 1 MP output, one image. Price basis: published per-megapixel list rate × 1 MP. Currency: USD. Not a Meituan direct tariff. fal.ai official model page, observed 2026-09-26.
Other ways to run it
- Run it as published · needs ~35.7 GB, NVIDIA only · from huggingface.co/meituan-longcat
uv run --with git+https://github.com/huggingface/diffusers --with transformers --with accelerate --with torch python -c "import torch; from PIL import Image; from diffusers import LongCatImageEditPipeline; p=LongCatImageEditPipeline.from_pretrained('meituan-longcat/LongCat-Image-Edit', torch_dtype=torch.bfloat16); p.enable_model_cpu_offload(); p(Image.open('input.png').convert('RGB'), 'turn the cat into a dog', negative_prompt='', guidance_scale=4.5, num_inference_steps=50, num_images_per_prompt=1, generator=torch.Generator('cpu').manual_seed(43)).images[0].save('edit_example.png')"
Weights and documentation: huggingface.co/meituan-longcat ↗
$ how the score was worked out · Edit-Compass & EditReward-Com… 39%, Meituan LongCat — LongCat-Ima… 26%, ChartSync: A Benchmark for Vi… 18% and 3 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 |
|---|---|---|
 Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling | 1 | 39% |
| Meituan LongCat — LongCat-Image technical report · lab-reported | 2 | 26% |
| ChartSync: A Benchmark for Visuo-Logical Cascading Chart Editing (arXiv:2607.10301v1) | 1 | 18% |
| GEditBench v2: A Human-Aligned Benchmark for General Image Editing | 1 | 8% |
| PaintBench: Deterministic Evaluation of Precise Visual Editing (arXiv:2606.00188v1) | 1 | 4% |
| ViGoR-Bench: How Far Are Visual Generative Models From Zero-Shot Visual Reasoners? (arXiv:2603.25823v1) · lab-reported | 1 | 3% |
| prior · pull toward the middle | 1% | |
| weighted average | = 56.4 |
not counted (2)
| Meituan LongCat — LongCat-Image technical report | GEdit-Bench-EN overall (G_O, GPT-4.1 judge) [correct] | 76.4% | fewer than three models have this test yet, or this source ran it for this model alone |
| Meituan LongCat — LongCat-Image technical report | ImgEdit-Bench overall (GPT-4.1 judge) [correct] | 90.0% | 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