SAVRN
Search Contact SAVRN

SAVRN Model Hub · Comparisons

siglip-so400m-patch14-384 vs siglip2-giant-opt-patch16-384

Siglip-so400m-patch14-384 has 878M parameters and siglip2-giant-opt-patch16-384 has 1.9B parameters; both are released under Apache License 2.0; at 16-bit, siglip-so400m-patch14-384 needs about 2.1 GB (1x MI300X from $1.85 an hour) and siglip2-giant-opt-patch16-384 about 4.5 GB (1x MI300X from $1.85 an hour).

Published metadata for 2 models, each read from its own repository.
Field siglip-so400m-patch14-384
google/siglip-so400m-patch14-384
siglip2-giant-opt-patch16-384
google/siglip2-giant-opt-patch16-384
Publisher Google Google
Task Zero shot image classification Zero shot image classification
Modality Other Other
Parameters, as reported 878M parameters 1.9B parameters
Architecture SiglipModel siglip
Library transformers transformers
Context length Not stated Not stated
Repository size 3.5 GB 7.5 GB
Artifact formats safetensors safetensors
License apache-2.0 apache-2.0
Access Open weights, no gate Open weights, no gate
Memory at 16-bit (weights and margin) 2.1 GB 4.5 GB
Cheapest GPUs at 16-bit, per hour 1x MI300X, $1.85 1x MI300X, $1.85
Memory at 4-bit (weights and margin) 0.5 GB 1.1 GB
Cheapest GPUs at 4-bit, per hour 1x MI300X, $1.85 1x MI300X, $1.85
Revision viewed 9fdffc58afc9 a713301b217d
Downloads reported by the hub 1.3M 2.4M
Last observed 2026-09-18 2026-09-18

An evaluation row appears only where at least two of these models report the same benchmark with the same stated configuration, metric, unit and setup. Different evaluators stay named in each cell. Values are shown as reported: no unit conversion, no ranking.

SAVRN's Notes on siglip-so400m-patch14-384

Two point one gigabytes. That is the whole running footprint of this SigLIP image-text model at 16-bit, and the cheapest slot we price it on is one MI300X with 192 GB at $1.85 an hour on demand. Nearly the whole card sits idle, so run it beside other work. Its job is zero-shot image classification at a 384 by 384 input, using a sigmoid loss instead of the usual CLIP objective; 8-bit needs 1.1 GB and 4-bit 0.5 GB.

Apache 2.0 permits commercial use, modification and redistribution with a patent grant, so product use is uncomplicated. The team that released SigLIP did not write its model card, so the three papers listed for it, arXiv:2303.15343, 2305.13035 and 2209.06794, are the primary reference. The weights are stored in float32, 3.5 GB across nine files, so budget the download at that size and convert down yourself.

SAVRN's Notes on siglip2-giant-opt-patch16-384

The download is 7.5 GB across 11 files, but the 16-bit copy needs 4.5 GB of memory and 4-bit needs 1.1 GB, so budget disk and memory separately for siglip2-giant-opt-patch16-384. Google publishes it for zero-shot image classification and image-text retrieval, and as the vision tower in front of a language model. On the cheapest slot we track, one MI300X with 192 GB at $1.85 an hour on demand, 4.5 GB is a sliver, so it rides beside the model it feeds.

No context length applies to an encoder, so the license carries the deployment questions: Apache License 2.0 permits commercial use, modification and redistribution, with notices kept and significant changes stated. What a buyer should check is the three papers the file cites, arXiv:2502.14786, arXiv:2303.15343 and arXiv:2209.06794, for the training recipe, and that the 1,152 hidden size and 256,000-entry vocabulary match the model it will feed.

Questions

Which is larger, siglip-so400m-patch14-384 or siglip2-giant-opt-patch16-384?

siglip2-giant-opt-patch16-384 (1.9B parameters) is larger than siglip-so400m-patch14-384 (878M parameters), by the parameter counts their publishers report.

Which is cheaper to run, siglip-so400m-patch14-384 or siglip2-giant-opt-patch16-384?

At 4-bit, siglip-so400m-patch14-384 fits on 1x MI300X from $1.85 an hour and siglip2-giant-opt-patch16-384 on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use siglip-so400m-patch14-384 commercially?

Yes. siglip-so400m-patch14-384 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

Can I use siglip2-giant-opt-patch16-384 commercially?

Yes. siglip2-giant-opt-patch16-384 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

Related Comparisons