Meta Vision Models

Compare all 16 Meta vision models we track, 13 of them open-weight. Try 7 of them on your own images, free in the Roboflow Playground.

16 models · 13 open-weight · 7 free to try · Updated Sep 2026

About Meta Vision Models

Meta ships vision models on two tiers, and our catalog tracks 16 of them. 13 are open-weight: the weights are downloadable, so you can self-host under their own licenses. The other 3 are proprietary, reached through an API and billed by the provider. The open-weight side runs on Apache 2.0, Custom, and MIT licenses. The newest additions are Muse Spark 1.3 (Sep 2026) and Muse Glimmer 30B (Aug 2026). The lineup reaches back to ResNet-34, released Dec 2015.

Between them, the Meta models we list cover 20 distinct vision tasks. The widest coverage is object detection (10 models), classification (9), and captioning (6). For object detection, start with Muse Spark 1.3, Muse Glimmer 30B, or Muse Spark 1.2.

On the open-weight tier, Llama 4 Maverick is the largest at 400B parameters and ResNet-32 the smallest at 0.46M, small enough to run on a single GPU or on-device. 9 of the 13 open Meta models carry a permissive license, so commercial use is straightforward; the other 4 ship under terms worth reading before you deploy.

On the API tier, Meta's most recent entry is Muse Spark 1.3 (Sep 2026), billed per token by the provider rather than run on your own hardware. Which tier to start on is a constraints question, not a quality one: reach for the open-weight side when you need offline inference, predictable per-image cost, or a checkpoint you can fine-tune on your own data, and for the API side when you want broad general reasoning without managing GPUs. 7 of the 16 Meta vision models run live in the Roboflow Playground, so you can run the same image through several of them and compare the answers before committing to one.

Which Meta Model Should You Use?

What each of the 16 Meta vision models in our catalog is built for, and how you run it.

Muse Spark 1.3: Best for: Object Detection and OCR. Proprietary: no downloadable weights, billed per token through Meta's API. Runnable in the Playground.
Muse Glimmer 30B: Best for: Object Detection and OCR. Open weights: Apache 2.0 license, 29.6B parameters. Self-host it or run it here. Runnable in the Playground.
Muse Spark 1.2: Best for: Object Detection and OCR. Proprietary: no downloadable weights, billed per token through Meta's API. Runnable in the Playground.
Muse Spark 1.1: Best for: Object Detection and OCR. Proprietary: no downloadable weights, billed per token through Meta's API. Runnable in the Playground.
SAM 3: Best for: Promptable Concept Segmentation and Zero Shot Segmentation. Open weights: Custom license. Self-host it or run it here. Runnable in the Playground.
SAM 3D Objects: Best for: Instance Segmentation and 3D Reconstruction. Open weights: Custom license. Self-host it or run it here.
Llama 4 Maverick: Best for: Object Detection and OCR. Open weights: Custom license, 400B parameters. Self-host it or run it here. Runnable in the Playground.
Llama 4 Scout: Best for: Object Detection and OCR. Open weights: Custom license, 109B parameters. Self-host it or run it here. Runnable in the Playground.
Segment Anything Model 2 (SAM 2): Best for: Instance Segmentation. Open weights: Apache 2.0 license, 38.9M-224.4M parameters. Self-host it or run it here. Small enough for on-device and edge deployment.
DINOv2: Best for: Image Embedding and Image Similarity. Open weights: Apache 2.0 license, 21M-1.1B parameters. Self-host it or run it here. Small enough for on-device and edge deployment.
Segment Anything Model (SAM): Best for: Instance Segmentation. Open weights: Apache 2.0 license, 91M-636M parameters. Self-host it or run it here. Small enough for on-device and edge deployment.
DETR: Best for: Object Detection. Open weights: Apache 2.0 license, ~41M parameters. Self-host it or run it here. Small enough for on-device and edge deployment.
Detectron2: Best for: Instance Segmentation and Semantic Segmentation. Open weights: Apache 2.0 license. Self-host it or run it here.
Mask R-CNN: Best for: Instance Segmentation and Object Detection. Open weights: MIT license, 44.4M parameters. Self-host it or run it here. Small enough for on-device and edge deployment.
ResNet-32: Best for: Classification. Open weights: MIT license, 0.46M parameters. Self-host it or run it here. Small enough for on-device and edge deployment.
ResNet-34: Best for: Classification. Open weights: MIT license, 21.8M parameters. Self-host it or run it here. Small enough for on-device and edge deployment.

Open-Source Meta Models

13 models with downloadable weights you can self-host under their licenses (Apache 2.0, Custom, and MIT). 4 run live in the Playground through hosted APIs, so self-hosting is optional.

				Actions
Muse Glimmer 30BMeta	29.6B	Apache 2.0	Aug 2026	Try
SAM 3Meta	—	Custom	Nov 2025	Try
SAM 3D ObjectsMeta	—	Custom	Nov 2025
Llama 4 MaverickMeta	400B	Custom	Apr 2025	Try
Llama 4 ScoutMeta	109B	Custom	Apr 2025	Try
Segment Anything Model 2 (SAM 2)Meta	38.9M-224.4M	Apache 2.0	Jul 2024
DINOv2Meta	21M-1.1B	Apache 2.0	Apr 2023
Segment Anything Model (SAM)Meta	91M-636M	Apache 2.0	Apr 2023
DETRMeta	~41M	Apache 2.0	May 2020
Detectron2Meta	—	Apache 2.0	Sep 2019
Mask R-CNNMeta	44.4M	MIT	Oct 2017
ResNet-32Meta	0.46M	MIT	Dec 2015
ResNet-34Meta	21.8M	MIT	Dec 2015

Meta Models via API

3 proprietary models where the weights aren't downloadable: access is through each provider's API and billed by them. Try all of them free in the Playground.

					Actions
Muse Spark 1.3Meta NEW	$1.25	$4.25	1.0M	Sep 2026	Try
Muse Spark 1.2Meta	$1.25	$4.25	1.0M	Aug 2026	Try
Muse Spark 1.1Meta	$1.25	$4.25	1.0M	Jul 2026	Try

Frequently Asked Questions About Meta Vision Models

Which Meta models can do object detection?

10 of the 16 Meta vision models we track handle object detection: Muse Spark 1.3, Muse Glimmer 30B, Muse Spark 1.2, Muse Spark 1.1, SAM 3, and Llama 4 Maverick, and 4 more. Each model page lists its full task coverage, license, and specs.

Which Meta models can do classification?

9 of the 16 Meta vision models we track handle classification: Muse Spark 1.3, Muse Glimmer 30B, Muse Spark 1.2, Muse Spark 1.1, Llama 4 Maverick, and Llama 4 Scout, and 3 more. Each model page lists its full task coverage, license, and specs.

Are Meta vision models open source?

Partly. 13 of the 16 Meta vision models we track are open-weight (for example Muse Glimmer 30B, SAM 3, and SAM 3D Objects), downloadable and self-hostable under their licenses (Apache 2.0, Custom, and MIT). The other 3 are proprietary and reached through Meta's API.

What is the best Meta model for object detection?

We do not publish a Meta-only ranking, so pick on constraints rather than a label. 10 Meta models handle object detection; the most recent is Muse Spark 1.3 (Sep 2026), and Muse Glimmer 30B is the newest open-weight option if you need to self-host. Measured scores across every lab we test are on our object detection benchmark, linked at the bottom of this page. For fixed categories in production, a model fine-tuned on your own data typically beats any general-purpose model.

How many Meta vision models are on Roboflow Playground?

We track 16 live Meta vision models, 13 open-weight and 3 available through an API. The most recent addition is Muse Spark 1.3, released Sep 2026.

Can I try Meta vision models for free?

Yes. 7 of the 16 Meta models run live in the Roboflow Playground. Upload your own image, run several models on it at once, and compare the outputs side by side. No setup and no account required.

This page lists all 16 Meta vision models in the Roboflow Playground catalog: 13 open-weight models you can self-host and 3 proprietary models accessed through an API. They cover object detection, classification, and captioning, among other tasks. 7 of them run live in the Roboflow Playground on your own images. Compare licenses, parameters, prices, and release dates side by side, or open any model page for full details.