Gallery uses Postgres as its search database for both metadata and contextual CLIP search.
Contextual CLIP search is powered by the VectorChord extension, utilizing machine learning models like CLIP to provide relevant search results. This allows for freeform searches without requiring specific keywords in the image or video metadata.
Advanced Search Filters
In addition, Gallery offers advanced search functionality, allowing you to find specific content using customizable search filters. These filters include location, one or more faces, specific albums, and more. You can try out the search filters on the Demo site.
The Search Palette also supports typed filter syntax, so you can apply filters from the keyboard with searches like beach person:anna from:2025 type:video. See Search Palette typed filter syntax for the supported keys and resolution behavior.
You can search the following types of content:
| Type | Description |
|---|
| People | Faces that are recognized in your photos/videos. |
| Contextual | Content of the photos and videos. |
| File name or extension | Full or partial file's name, or file's extension |
| Description | Description added to assets. |
| Optical Character Recognition (OCR) | Text in images |
| Locations | Cities, states, and countries from reverse geocoding. |
| Tags | Tags assigned or extracted from assets. |
| Camera | make, model and lens model |
| Time frame | Start and end date of a specific time bucket |
| Media type | Image or video or both |
| Display options | In Archive, in Favorites or Not in any album |
| Star rating | Minimum user-assigned star rating |
Configuration
Navigating to Administration > Settings > Machine Learning Settings > Smart Search will show the options available.
CLIP models
The default search model is fast, but there are many other options that can provide better search results. The tradeoff of using these models is that they're slower and/or use more memory (both when indexing images with background Smart Search jobs and when searching).
The first step of choosing the right model for you is to know which languages your users will search in.
If your users will only search in English, then the CLIP section is the first place to look. This is a curated list of the models that generally perform the best for their size class. The models here are ordered from higher to lower quality. This means that the top models will generally rank the most relevant results higher and have a higher capacity to understand descriptive, detailed, and/or niche queries. The models are also generally ordered from larger to smaller, so consider the impact on memory usage, job processing and search speed when deciding on one. The smaller models in this list are not too different in quality and many times faster.
Multilingual models are also available so users can search in their native language. Use these models if you expect non-English searches to be common. They can be separated into two search patterns:
nllb models expect the search query to be in the language specified in the user settingsxlm and siglip2 models understand search text regardless of the current language setting
nllb models tend to perform the best and are recommended when users primarily searches in their native, non-English language. xlm and siglip2 models are more flexible and are recommended for mixed language search, where the same user might search in different languages at different times.
For more details, check the tables below to see how they compare in memory usage, speed and quality by language.
Once you've chosen a model, follow these steps:
- Copy the name of the model (e.g.
ViT-B-16-SigLIP__webli)
- Go to the Smart Search settings
- Paste the model name into the Model Name section
- Save the settings
- Go to the Job Status page
- Click "All" next to "Smart Search" to begin re-processing your assets with the new model
- (Optional) Confirm that the logs for the server and machine learning service don't have relevant errors
In rare instances, changing the model might leave bits of the old model's incompatible data in the database, causing errors when processing Smart Search jobs. If you notice errors like this in the logs, you can change the model back to the previous one and save, then repeat steps 3-7.
Please note that memory and execution time values are only estimates: actual usage will be different depending on many factors. As such, it's mainly intended as a way to compare the relative tradeoffs of each model.
Reference
Memory and execution time estimates were obtained without acceleration on a 7800x3D processor running bare metal Linux. All testing and evaluation was done at f32 precision (the default in Gallery).
Execution Time (ms): After warming up the model with one pass, the mean execution time of 100 passes with the same input.
Memory (MiB): The peak RSS usage of the process after performing the above timing benchmark. Does not include image decoding, concurrent processing, the web server, etc., which are relatively constant factors.
Recall (%): Evaluated on Crossmodal-3600, the average of the recall@1, recall@5 and recall@10 results for zeroshot image retrieval. Chinese (Simplified), English, French, German, Italian, Japanese, Korean, Polish, Russian, Spanish and Turkish are additionally tested on XTD-10. Chinese (Simplified) and English are additionally tested on Flickr30k. The recall metrics are the average across all tested datasets.
Pareto Optimal: Whether the model is not completely outclassed by another model. Try to use models that are optimal for the languages relevant to you. Specifically, for a given model and language, if there's another model that's better for that language in at least one respect (memory usage, execution time, recall) while being at least as good for that language in every other way, then the model is not optimal for that language.
English
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 85.99 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 85.96 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 85.96 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 85.93 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 85.78 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 85.75 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 85.62 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 85.53 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 85.48 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 85.47 | ✅ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 85.09 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 85.03 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 84.86 | ✅ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 84.61 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 84.17 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 83.51 | ❌ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 83.28 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 83.24 | ❌ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 83.23 | ❌ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 83.19 | ✅ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 82.54 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 82.43 | ❌ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 82.36 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 82.28 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 81.9 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 81.9 | ❌ |
| ViT-L-14__laion2b-s32b-b82k | 2233 | 20.56 | 80.82 | ❌ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 80.65 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 80.16 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 79.78 | ❌ |
| ViT-L-14__laion400m_e31 | 2183 | 19.87 | 78.64 | ❌ |
| ViT-L-14__laion400m_e32 | 2218 | 19.73 | 78.6 | ❌ |
| ViT-B-16-plus-240__laion400m_e32 | 1246 | 6.95 | 78.06 | ❌ |
| ViT-B-16-plus-240__laion400m_e31 | 1263 | 6.94 | 78.06 | ❌ |
| ViT-B-32__laion2b-s34b-b79k | 1001 | 2.29 | 77.62 | ✅ |
| ViT-B-32__laion2b_e16 | 1004 | 2.38 | 77.47 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 76.91 | ❌ |
| ViT-B-16__laion400m_e32 | 975 | 4.98 | 76.43 | ✅ |
| ViT-B-16__laion400m_e31 | 991 | 5.04 | 76.35 | ❌ |
| ViT-B-32__laion400m_e31 | 999 | 2.28 | 73.83 | ✅ |
| ViT-B-32__laion400m_e32 | 1003 | 2.35 | 73.62 | ❌ |
| RN50x64__openai | 5079 | 48.79 | 73.34 | ❌ |
| ViT-L-14__openai | 2212 | 19.91 | 72.99 | ❌ |
| ViT-L-14-336__openai | 2616 | 43.45 | 72.76 | ❌ |
| RN50x16__openai | 2221 | 15.87 | 72.59 | ❌ |
| RN50x4__openai | 1416 | 5.85 | 70.8 | ❌ |
| ViT-B-16__openai | 985 | 5.03 | 70.01 | ❌ |
| ViT-B-32__openai | 1004 | 2.26 | 69.9 | ✅ |
| RN101__openai | 1111 | 3.21 | 69.3 | ❌ |
| RN50__openai | 913 | 2.39 | 69.02 | ✅ |
| RN50__cc12m | 914 | 2.37 | 64.59 | ✅ |
| RN101__yfcc15m | 1111 | 3.22 | 55.21 | ❌ |
| RN50__yfcc15m | 908 | 2.34 | 53.63 | ✅ |
Arabic
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 77.3 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 76.44 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 74.03 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 73.19 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 69.31 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 69.29 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 69.29 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 68.64 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 68.35 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 68.25 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 68.23 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 67.56 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 67.28 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 66.89 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 66.52 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 64.1 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 61.71 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 60.7 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 59.66 | ✅ |
Bengali
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 76.16 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 75.83 | ❌ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 73.75 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 73.34 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 36.43 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 26.56 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 26.54 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 26.19 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 26.19 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 25.92 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 25.15 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 24.18 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 21.44 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 21.11 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 20.94 | ✅ |
Chinese (Simplified)
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 79.7 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 78.94 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 75.22 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 74.8 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 73.91 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 72.8 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 72.77 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 72.41 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 72.36 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 71.59 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 71.37 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 71.3 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 71.11 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 70.95 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 70.51 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 67.48 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 66.84 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 65.7 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 63.38 | ❌ |
Croatian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 87.46 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 87.19 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 82.98 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 82.92 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 81.93 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 73.77 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 73.21 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 73.2 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 72.95 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 72.89 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 72.88 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 72.85 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 72.69 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 70.73 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 70.45 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 70.43 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 69.97 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 54.31 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 53.3 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 35.64 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 35.17 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 33.65 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 33.55 | ❌ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 20.05 | ✅ |
Cusco Quechua
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 38.08 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 37.87 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 33.41 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 33.06 | ✅ |
Czech
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 73.76 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 71.57 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 69.86 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 67.49 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 67.15 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 63.62 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 63.35 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 63.09 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 63.07 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 62.98 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 62.82 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 62.73 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 62.29 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 62.12 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 61.74 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 61.52 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 61.01 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 54.81 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 54.31 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 33.58 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 33.48 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 32.38 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 32.32 | ❌ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 22.89 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 22.66 | ❌ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 22.6 | ❌ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 22.25 | ❌ |
Danish
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 87.16 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 86.88 | ❌ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 84.18 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 84.03 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 83.75 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 83.32 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 83.25 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 82.3 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 82.19 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 81.87 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 81.44 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 81.42 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 80.0 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 79.82 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 79.08 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 75.07 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 74.84 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 67.68 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 67.2 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 65.59 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 65.36 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 42.31 | ✅ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 41.46 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 40.52 | ❌ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 31.31 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 30.97 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 30.87 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 30.51 | ❌ |
Dutch
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 80.05 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 79.81 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 79.72 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 79.72 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 79.64 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 79.49 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 79.41 | ❌ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 79.31 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 78.92 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 78.48 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 78.22 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 78.0 | ✅ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 77.22 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 76.69 | ❌ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 75.94 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 75.6 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 75.33 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 75.04 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 72.97 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 72.72 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 72.06 | ✅ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 72.06 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 70.81 | ✅ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 69.82 | ✅ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 67.54 | ❌ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 66.77 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 66.6 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 65.67 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 65.29 | ✅ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 41.1 | ❌ |
| ViT-L-14__laion2b-s32b-b82k | 2233 | 20.56 | 34.29 | ❌ |
| ViT-L-14__laion400m_e31 | 2183 | 19.87 | 29.65 | ❌ |
| ViT-L-14__laion400m_e32 | 2218 | 19.73 | 29.56 | ❌ |
| ViT-B-32__laion2b-s34b-b79k | 1001 | 2.29 | 29.54 | ✅ |
| ViT-B-32__laion2b_e16 | 1004 | 2.38 | 29.36 | ❌ |
| ViT-B-16-plus-240__laion400m_e32 | 1246 | 6.95 | 27.76 | ❌ |
| ViT-B-16-plus-240__laion400m_e31 | 1263 | 6.94 | 27.76 | ❌ |
| ViT-B-16__laion400m_e32 | 975 | 4.98 | 25.67 | ✅ |
| ViT-B-32__laion400m_e32 | 1003 | 2.35 | 25.59 | ❌ |
| ViT-B-16__laion400m_e31 | 991 | 5.04 | 25.53 | ❌ |
| ViT-B-32__laion400m_e31 | 999 | 2.28 | 25.52 | ✅ |
| ViT-L-14__openai | 2212 | 19.91 | 22.31 | ❌ |
| RN50x64__openai | 5079 | 48.79 | 22.27 | ❌ |
| ViT-L-14-336__openai | 2616 | 43.45 | 21.8 | ❌ |
| RN50x16__openai | 2221 | 15.87 | 20.69 | ❌ |
Filipino
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 67.57 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 65.64 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 61.21 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 59.42 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 36.81 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 35.72 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 34.75 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 34.63 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 34.39 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 34.27 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 34.14 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 33.98 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 30.57 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 30.57 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 30.05 | ✅ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 24.92 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 24.02 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 23.37 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 22.69 | ✅ |
Finnish
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 84.27 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 83.93 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 79.41 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 78.94 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 75.49 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 63.46 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 63.16 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 63.08 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 63.03 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 62.28 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 61.92 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 61.81 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 61.76 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 61.05 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 57.8 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 57.69 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 57.05 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 40.26 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 40.06 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 31.75 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 31.74 | ❌ |
French
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 86.5 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 86.5 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 86.39 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 86.15 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 86.1 | ❌ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 86.07 | ❌ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 86.06 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 85.89 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 85.67 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 85.67 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 85.63 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 85.39 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 85.35 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 84.97 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 83.8 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 82.96 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 82.91 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 82.52 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 81.21 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 80.23 | ✅ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 79.85 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 79.47 | ✅ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 79.3 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 77.49 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 76.82 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 75.94 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 75.3 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 75.24 | ❌ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 69.33 | ❌ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 64.41 | ❌ |
| ViT-L-14__laion2b-s32b-b82k | 2233 | 20.56 | 62.86 | ❌ |
| ViT-L-14__laion400m_e32 | 2218 | 19.73 | 59.27 | ❌ |
| ViT-L-14__laion400m_e31 | 2183 | 19.87 | 59.09 | ❌ |
| ViT-B-16-plus-240__laion400m_e32 | 1246 | 6.95 | 58.25 | ❌ |
| ViT-B-16-plus-240__laion400m_e31 | 1263 | 6.94 | 58.25 | ❌ |
| ViT-B-32__laion2b_e16 | 1004 | 2.38 | 56.97 | ✅ |
| ViT-B-32__laion2b-s34b-b79k | 1001 | 2.29 | 56.21 | ✅ |
| ViT-B-32__laion400m_e31 | 999 | 2.28 | 53.36 | ✅ |
| ViT-B-16__laion400m_e32 | 975 | 4.98 | 53.33 | ✅ |
| ViT-B-16__laion400m_e31 | 991 | 5.04 | 53.26 | ❌ |
| ViT-B-32__laion400m_e32 | 1003 | 2.35 | 53.22 | ❌ |
| ViT-L-14__openai | 2212 | 19.91 | 46.34 | ❌ |
| RN50x64__openai | 5079 | 48.79 | 46.3 | ❌ |
| ViT-L-14-336__openai | 2616 | 43.45 | 45.95 | ❌ |
| RN50x16__openai | 2221 | 15.87 | 45.69 | ❌ |
| RN50x4__openai | 1416 | 5.85 | 42.48 | ❌ |
| RN101__openai | 1111 | 3.21 | 40.16 | ❌ |
| ViT-B-16__openai | 985 | 5.03 | 40.1 | ❌ |
| ViT-B-32__openai | 1004 | 2.26 | 38.27 | ✅ |
| RN50__openai | 913 | 2.39 | 37.8 | ✅ |
German
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 87.32 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 87.29 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 87.29 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 87.21 | ✅ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 87.18 | ❌ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 87.14 | ❌ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 87.07 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 86.83 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 86.81 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 86.75 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 86.74 | ✅ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 86.68 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 86.56 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 86.16 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 84.54 | ❌ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 84.41 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 84.25 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 83.8 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 82.59 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 81.53 | ✅ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 81.34 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 81.15 | ✅ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 81.05 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 78.35 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 76.56 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 76.0 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 75.21 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 75.14 | ❌ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 65.86 | ❌ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 56.87 | ❌ |
| ViT-L-14__laion2b-s32b-b82k | 2233 | 20.56 | 47.19 | ❌ |
| ViT-L-14__laion400m_e32 | 2218 | 19.73 | 43.36 | ❌ |
| ViT-L-14__laion400m_e31 | 2183 | 19.87 | 43.0 | ❌ |
| ViT-B-32__laion2b_e16 | 1004 | 2.38 | 41.81 | ✅ |
| ViT-B-32__laion2b-s34b-b79k | 1001 | 2.29 | 40.43 | ✅ |
| ViT-B-16-plus-240__laion400m_e32 | 1246 | 6.95 | 40.41 | ❌ |
| ViT-B-16-plus-240__laion400m_e31 | 1263 | 6.94 | 40.41 | ❌ |
| ViT-B-16__laion400m_e31 | 991 | 5.04 | 37.71 | ✅ |
| ViT-B-16__laion400m_e32 | 975 | 4.98 | 37.64 | ✅ |
| ViT-B-32__laion400m_e31 | 999 | 2.28 | 36.04 | ✅ |
| ViT-B-32__laion400m_e32 | 1003 | 2.35 | 35.9 | ❌ |
| RN50x64__openai | 5079 | 48.79 | 34.19 | ❌ |
| ViT-L-14__openai | 2212 | 19.91 | 33.1 | ❌ |
| ViT-L-14-336__openai | 2616 | 43.45 | 32.25 | ❌ |
| RN50x16__openai | 2221 | 15.87 | 30.56 | ❌ |
| RN50x4__openai | 1416 | 5.85 | 29.2 | ❌ |
| ViT-B-16__openai | 985 | 5.03 | 25.77 | ❌ |
| RN101__openai | 1111 | 3.21 | 25.46 | ❌ |
| RN50__openai | 913 | 2.39 | 24.92 | ✅ |
| ViT-B-32__openai | 1004 | 2.26 | 24.13 | ✅ |
Greek
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 74.58 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 73.28 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 71.28 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 69.16 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 68.21 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 64.69 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 61.64 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 61.03 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 60.63 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 60.41 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 60.1 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 60.06 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 60.06 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 59.44 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 59.44 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 59.43 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 58.78 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 53.42 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 53.24 | ❌ |
Hebrew
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 88.04 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 87.09 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 83.93 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 83.84 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 80.78 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 74.59 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 72.73 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 72.25 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 72.19 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 72.15 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 72.08 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 72.07 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 72.06 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 71.78 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 70.55 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 70.03 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 69.34 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 60.33 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 58.49 | ❌ |
Hindi
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 62.02 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 61.67 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 58.68 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 58.54 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 38.54 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 36.95 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 36.62 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 36.06 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 35.76 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 35.34 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 35.17 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 34.94 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 34.91 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 34.19 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 33.56 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 32.06 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 31.85 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 27.87 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 27.08 | ❌ |
Hungarian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 85.59 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 85.25 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 81.74 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 80.34 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 80.14 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 74.94 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 74.2 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 74.03 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 73.96 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 73.95 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 73.9 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 73.59 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 73.12 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 72.5 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 72.33 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 71.83 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 70.57 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 58.31 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 56.74 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 38.26 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 37.97 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 28.75 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 28.26 | ❌ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 24.88 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 24.39 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 24.29 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 24.16 | ❌ |
Indonesian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 85.46 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 85.12 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 85.01 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 84.99 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 84.65 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 84.62 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 84.58 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 84.11 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 84.1 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 84.06 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 83.69 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 83.61 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 82.31 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 81.97 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 80.93 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 79.84 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 77.12 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 77.02 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 74.15 | ✅ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 71.44 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 69.94 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 65.87 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 65.19 | ❌ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 59.95 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 59.38 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 57.88 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 57.52 | ✅ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 54.11 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 50.02 | ❌ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 23.25 | ❌ |
Italian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 87.17 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 86.91 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 86.83 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 86.77 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 86.67 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 86.42 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 86.35 | ✅ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 86.34 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 86.18 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 86.17 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 85.84 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 85.8 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 85.7 | ✅ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 85.67 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 83.32 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 82.95 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 82.73 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 82.72 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 81.07 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 80.8 | ✅ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 80.6 | ✅ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 80.35 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 78.79 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 76.62 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 76.51 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 76.08 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 75.29 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 75.29 | ❌ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 74.84 | ❌ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 56.32 | ❌ |
| ViT-L-14__laion2b-s32b-b82k | 2233 | 20.56 | 47.25 | ❌ |
| ViT-L-14__laion400m_e32 | 2218 | 19.73 | 43.09 | ❌ |
| ViT-L-14__laion400m_e31 | 2183 | 19.87 | 42.99 | ❌ |
| ViT-B-16-plus-240__laion400m_e32 | 1246 | 6.95 | 40.29 | ❌ |
| ViT-B-16-plus-240__laion400m_e31 | 1263 | 6.94 | 40.29 | ❌ |
| ViT-B-32__laion2b_e16 | 1004 | 2.38 | 39.67 | ✅ |
| ViT-B-32__laion2b-s34b-b79k | 1001 | 2.29 | 39.03 | ✅ |
| ViT-B-16__laion400m_e32 | 975 | 4.98 | 36.14 | ✅ |
| ViT-B-16__laion400m_e31 | 991 | 5.04 | 35.89 | ❌ |
| ViT-B-32__laion400m_e32 | 1003 | 2.35 | 35.59 | ❌ |
| ViT-B-32__laion400m_e31 | 999 | 2.28 | 35.56 | ✅ |
| RN50x64__openai | 5079 | 48.79 | 33.53 | ❌ |
| ViT-L-14__openai | 2212 | 19.91 | 32.19 | ❌ |
| ViT-L-14-336__openai | 2616 | 43.45 | 30.95 | ❌ |
| RN50x16__openai | 2221 | 15.87 | 28.85 | ❌ |
| RN50x4__openai | 1416 | 5.85 | 25.75 | ❌ |
| ViT-B-16__openai | 985 | 5.03 | 25.18 | ❌ |
| RN101__openai | 1111 | 3.21 | 24.48 | ❌ |
| RN50__openai | 913 | 2.39 | 23.89 | ✅ |
| ViT-B-32__openai | 1004 | 2.26 | 23.39 | ✅ |
Japanese
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 83.95 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 82.21 | ❌ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 81.55 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 78.72 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 78.53 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 75.93 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 66.86 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 65.59 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 65.48 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 65.36 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 64.47 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 64.17 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 64.08 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 63.69 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 63.33 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 63.02 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 58.39 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 56.38 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 53.16 | ❌ |
Korean
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 80.56 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 80.53 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 77.09 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 77.08 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 76.97 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 76.92 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 76.58 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 76.2 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 75.95 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 75.86 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 75.67 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 75.49 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 74.6 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 74.52 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 73.88 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 71.09 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 68.87 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 67.94 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 66.39 | ✅ |
Maori
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 48.43 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 46.12 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 42.8 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 40.85 | ✅ |
Norwegian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 81.36 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 80.96 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 77.65 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 76.39 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 75.97 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 75.44 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 75.31 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 75.0 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 74.96 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 74.92 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 74.44 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 74.37 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 73.11 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 72.63 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 71.71 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 67.81 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 65.55 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 62.56 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 60.94 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 59.62 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 59.49 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 46.3 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 45.75 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 42.55 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 35.33 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 35.01 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 34.94 | ❌ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 34.39 | ❌ |
Persian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 79.52 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 78.99 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 76.32 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 76.3 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 76.11 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 75.56 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 75.38 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 74.92 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 74.86 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 74.73 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 74.32 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 74.31 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 73.42 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 72.56 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 71.9 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 69.79 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 68.55 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 68.26 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 65.16 | ❌ |
Polish
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 83.49 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 83.45 | ❌ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 83.11 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 82.99 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 82.96 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 82.93 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 82.61 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 82.26 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 82.24 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 82.03 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 82.03 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 81.92 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 81.27 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 80.0 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 79.65 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 76.75 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 76.52 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 75.1 | ✅ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 73.9 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 65.03 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 64.89 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 51.6 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 51.29 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 46.15 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 41.55 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 41.17 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 40.9 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 40.76 | ✅ |
Portuguese
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 82.12 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 81.84 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 81.69 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 81.69 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 81.54 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 81.39 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 80.56 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 80.34 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 80.02 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 79.99 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 79.93 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 79.61 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 79.12 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 78.87 | ❌ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 78.85 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 77.54 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 75.31 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 75.26 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 74.82 | ✅ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 74.48 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 74.47 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 73.92 | ✅ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 73.58 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 73.02 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 71.44 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 71.16 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 69.69 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 69.32 | ✅ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 59.86 | ❌ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 45.49 | ❌ |
| ViT-L-14__laion2b-s32b-b82k | 2233 | 20.56 | 37.86 | ❌ |
| ViT-L-14__laion400m_e32 | 2218 | 19.73 | 36.01 | ❌ |
| ViT-L-14__laion400m_e31 | 2183 | 19.87 | 35.75 | ❌ |
| ViT-B-16-plus-240__laion400m_e32 | 1246 | 6.95 | 33.25 | ❌ |
| ViT-B-16-plus-240__laion400m_e31 | 1263 | 6.94 | 33.25 | ❌ |
| ViT-B-32__laion2b_e16 | 1004 | 2.38 | 32.83 | ✅ |
| ViT-B-32__laion2b-s34b-b79k | 1001 | 2.29 | 32.62 | ✅ |
| ViT-B-32__laion400m_e32 | 1003 | 2.35 | 30.86 | ❌ |
| ViT-B-32__laion400m_e31 | 999 | 2.28 | 30.8 | ✅ |
| RN50x64__openai | 5079 | 48.79 | 30.58 | ❌ |
| ViT-B-16__laion400m_e32 | 975 | 4.98 | 30.18 | ✅ |
| ViT-B-16__laion400m_e31 | 991 | 5.04 | 29.93 | ❌ |
| ViT-L-14__openai | 2212 | 19.91 | 28.88 | ❌ |
| ViT-L-14-336__openai | 2616 | 43.45 | 28.49 | ❌ |
| RN50x16__openai | 2221 | 15.87 | 23.9 | ❌ |
| RN50x4__openai | 1416 | 5.85 | 22.94 | ❌ |
| ViT-B-16__openai | 985 | 5.03 | 22.55 | ❌ |
| RN50__openai | 913 | 2.39 | 21.85 | ✅ |
| ViT-B-32__openai | 1004 | 2.26 | 21.3 | ✅ |
| RN101__openai | 1111 | 3.21 | 21.14 | ❌ |
Romanian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 89.38 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 88.86 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 85.37 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 84.92 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 84.49 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 77.92 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 74.98 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 74.33 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 74.05 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 74.03 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 73.94 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 73.27 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 73.22 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 72.91 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 72.43 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 71.93 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 71.5 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 58.28 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 56.54 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 56.12 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 55.53 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 34.96 | ✅ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 26.33 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 26.05 | ❌ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 21.32 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 21.04 | ❌ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 20.76 | ❌ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 20.56 | ✅ |
Russian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 84.54 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 84.41 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 84.36 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 84.31 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 84.22 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 83.9 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 83.69 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 83.5 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 83.31 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 83.21 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 83.11 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 82.7 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 82.69 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 80.91 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 79.75 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 79.35 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 78.91 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 78.06 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 76.44 | ✅ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 42.81 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 42.1 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 24.95 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 24.25 | ❌ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 20.85 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 20.44 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 20.41 | ❌ |
Spanish
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 85.47 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 85.44 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 85.32 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 85.22 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 85.15 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 84.81 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 84.68 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 84.6 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 84.55 | ✅ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 84.27 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 84.15 | ✅ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 83.87 | ❌ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 83.74 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 83.61 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 83.15 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 81.7 | ❌ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 80.91 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 80.73 | ✅ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 80.69 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 80.3 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 79.8 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 79.71 | ✅ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 79.64 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 78.0 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 77.83 | ❌ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 76.87 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 76.66 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 75.99 | ✅ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 71.96 | ❌ |
| ViT-H-14__laion2b-s32b-b79k | 4676 | 39.06 | 62.06 | ❌ |
| ViT-L-14__laion2b-s32b-b82k | 2233 | 20.56 | 53.78 | ❌ |
| ViT-L-14__laion400m_e32 | 2218 | 19.73 | 50.13 | ❌ |
| ViT-L-14__laion400m_e31 | 2183 | 19.87 | 50.0 | ❌ |
| ViT-B-16-plus-240__laion400m_e32 | 1246 | 6.95 | 47.39 | ❌ |
| ViT-B-16-plus-240__laion400m_e31 | 1263 | 6.94 | 47.39 | ❌ |
| ViT-B-32__laion2b_e16 | 1004 | 2.38 | 46.47 | ✅ |
| ViT-B-32__laion2b-s34b-b79k | 1001 | 2.29 | 45.68 | ✅ |
| ViT-B-16__laion400m_e31 | 991 | 5.04 | 44.0 | ✅ |
| ViT-B-16__laion400m_e32 | 975 | 4.98 | 43.98 | ✅ |
| ViT-B-32__laion400m_e32 | 1003 | 2.35 | 43.8 | ❌ |
| ViT-B-32__laion400m_e31 | 999 | 2.28 | 43.73 | ✅ |
| RN50x64__openai | 5079 | 48.79 | 43.01 | ❌ |
| ViT-L-14__openai | 2212 | 19.91 | 42.96 | ❌ |
| ViT-L-14-336__openai | 2616 | 43.45 | 41.67 | ❌ |
| RN50x16__openai | 2221 | 15.87 | 40.21 | ❌ |
| RN50x4__openai | 1416 | 5.85 | 36.06 | ❌ |
| ViT-B-16__openai | 985 | 5.03 | 35.67 | ❌ |
| RN101__openai | 1111 | 3.21 | 34.62 | ❌ |
| ViT-B-32__openai | 1004 | 2.26 | 32.6 | ✅ |
| RN50__openai | 913 | 2.39 | 31.79 | ✅ |
Swahili
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 69.51 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 68.44 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 66.09 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 63.98 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 21.64 | ✅ |
Swedish
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 77.12 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 76.37 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 73.41 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 72.83 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 72.51 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 72.2 | ❌ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 72.1 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 72.06 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 71.84 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 71.7 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 71.7 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 71.61 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 71.51 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 71.45 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 71.23 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 67.48 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 66.93 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 66.37 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 64.86 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 62.35 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 61.51 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 56.74 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 55.92 | ❌ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 48.5 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 48.38 | ✅ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 48.06 | ❌ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 47.99 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 47.93 | ❌ |
| ViT-SO400M-14-SigLIP-384__webli | 4417 | 72.19 | 29.98 | ❌ |
Telugu
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 64.32 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 62.34 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 60.72 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 58.8 | ✅ |
Thai
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 79.99 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 79.07 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 76.13 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 75.23 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 74.04 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 66.03 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 45.87 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 45.69 | ❌ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 45.52 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 44.96 | ❌ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 44.75 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 44.66 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 43.99 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 43.91 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 43.06 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 41.86 | ❌ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 41.1 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 37.35 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 35.28 | ❌ |
Turkish
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 83.91 | ✅ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 83.74 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 81.26 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 80.21 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 79.34 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 79.22 | ✅ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 78.9 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 78.85 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 78.29 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 78.27 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 78.0 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 77.81 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 77.67 | ✅ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 77.33 | ✅ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 76.42 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 72.44 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 69.84 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 69.83 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 67.13 | ❌ |
| ViT-H-14-378-quickgelu__dfn5b | 5049 | 108.4 | 44.43 | ❌ |
| ViT-H-14-quickgelu__dfn5b | 4701 | 38.74 | 43.87 | ❌ |
| ViT-L-16-SigLIP-384__webli | 3396 | 47.6 | 35.1 | ❌ |
| ViT-L-16-SigLIP-256__webli | 3160 | 23.84 | 34.92 | ❌ |
| ViT-L-14-quickgelu__dfn2b | 2212 | 20.49 | 25.2 | ✅ |
| ViT-B-16-SigLIP-512__webli | 1828 | 26.17 | 24.55 | ✅ |
| ViT-B-16-SigLIP__webli | 1081 | 5.77 | 24.13 | ✅ |
| ViT-B-16-SigLIP-384__webli | 1128 | 13.53 | 24.08 | ❌ |
| ViT-B-16-SigLIP-256__webli | 1102 | 7.11 | 23.95 | ❌ |
Ukrainian
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 83.92 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 83.88 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 83.2 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 79.99 | ✅ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 79.31 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 78.73 | ✅ |
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 78.33 | ✅ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 77.95 | ❌ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 77.56 | ✅ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 77.49 | ✅ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 77.02 | ❌ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 76.87 | ❌ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 76.31 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 75.91 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 75.75 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 75.1 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 73.3 | ✅ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 65.28 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 63.95 | ❌ |
Vietnamese
| Model | Memory (MiB) | Execution Time (ms) | Recall (%) | Pareto Optimal |
|---|
| ViT-SO400M-16-SigLIP2-384__webli | 3854 | 56.57 | 85.86 | ✅ |
| ViT-SO400M-14-SigLIP2-378__webli | 3940 | 72.25 | 85.73 | ❌ |
| ViT-SO400M-16-SigLIP2-512__webli | 4050 | 107.67 | 85.67 | ❌ |
| ViT-gopt-16-SigLIP2-384__webli | 6585 | 146.84 | 85.5 | ❌ |
| ViT-L-16-SigLIP2-384__webli | 3057 | 51.7 | 84.93 | ✅ |
| ViT-SO400M-16-SigLIP2-256__webli | 3611 | 27.84 | 84.84 | ✅ |
| ViT-L-16-SigLIP2-512__webli | 3358 | 92.59 | 84.78 | ❌ |
| ViT-SO400M-14-SigLIP2__webli | 3622 | 27.63 | 84.34 | ✅ |
| ViT-gopt-16-SigLIP2-256__webli | 6475 | 64.51 | 84.33 | ❌ |
| ViT-L-16-SigLIP2-256__webli | 2830 | 23.77 | 83.93 | ✅ |
| nllb-clip-large-siglip__mrl | 4248 | 75.44 | 83.69 | ❌ |
| nllb-clip-large-siglip__v1 | 4226 | 75.05 | 83.19 | ❌ |
| XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k | 4014 | 39.14 | 81.88 | ❌ |
| ViT-B-16-SigLIP2__webli | 3038 | 5.81 | 80.88 | ✅ |
| nllb-clip-base-siglip__mrl | 4696 | 16.95 | 79.79 | ❌ |
| nllb-clip-base-siglip__v1 | 4675 | 15.17 | 79.38 | ❌ |
| ViT-B-32-SigLIP2-256__webli | 3061 | 3.31 | 77.73 | ✅ |
| XLM-Roberta-Base-ViT-B-32__laion5b_s13b_b90k | 3030 | 3.2 | 75.18 | ✅ |
| ViT-B-16-SigLIP-i18n-256__webli | 3029 | 6.87 | 73.05 | ✅ |
Feel free to make a feature request if there's a model you want to use that we don't currently support.
Relevance threshold
When combining a text search with metadata filters (e.g., searching "forest" and filtering by a specific country), the results may include photos that match the filter but have little visual similarity to the search term. This happens because the search returns all photos matching the filter, ranked by similarity — even when the best match in that filtered set is poor.
The max search distance setting adds a hard cutoff: results with a cosine distance above the threshold are excluded, regardless of how many remain. If no results pass the threshold, the search returns zero results rather than irrelevant ones.
Configuration
Set machineLearning.clip.maxDistance in Administration > Machine Learning > Smart Search, or in your config file. The value is a cosine distance (0 = identical, 2 = opposite).
| Value | Behavior |
|---|
0 | Disabled (default). All results returned regardless of similarity. |
0.5 | Very strict. Only strong visual matches. May miss borderline results. |
0.75 | Recommended starting point. Good balance of relevance and recall. |
1.0 | Permissive. Includes weaker matches. Useful for broad/abstract queries. |
Tuning tips
- Start at 0.75 and adjust based on your results. Lower values are stricter. If searches return nothing at 0.75, raise the threshold in small steps until relevant photos reappear — some libraries and CLIP models produce higher distances than others.
- Small changes can have a large effect. CLIP embeddings tend to cluster in a narrow distance range rather than being spread evenly. This means a threshold change from, say, 0.75 to 0.80 may dramatically increase the number of results. This is normal — not a bug.
- Different CLIP models produce different distance distributions. If you change your CLIP model, you may need to re-tune the threshold.
- Text-to-image vs. image-to-image searches have different distance characteristics. Text queries typically produce higher distances (looser matches) than searching by a similar photo. If you use both search modes, pick a threshold that works for text queries — it will be permissive enough for image-based searches too.