Compare the best 6 promptable concept segmentation models and try them on your own images, free in the Roboflow Playground.
6 models · 1 open-weight · 6 free to try · prices synced Oct 16, 2026
We haven't benchmarked promptable concept segmentation yet; scores and rankings appear on this site only where we've measured them. Until then, compare the models below, and for fixed categories in production expect a model fine-tuned on your own data to win.
1 model with downloadable weights you can self-host under their licenses (Custom). All run live in the Playground through hosted APIs, so self-hosting is optional.
5 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 | |||||
|---|---|---|---|---|---|
Gemini 4 ArgonGoogle NEW | $2.00 | $10.00 | 1M | Oct 2026 | |
GPT-6.1 SolOpenAI | $2.00 | $10.00 | 1.1M | Sep 2026 | |
GPT-6 LunaOpenAI | $0.10 | $0.50 | 1.1M | Sep 2026 | |
GPT-6 SolOpenAI | $2.00 | $10.00 | 1.1M | Sep 2026 | |
GPT-6 AstraOpenAI | $10.00 | $50.00 | 1.1M | Sep 2026 |
Concept segmentation extends promptable segmentation from "whatever is at this click" to "every instance of this concept": one prompt, instance masks for all matches. It is the newest family on this site, popularized by SAM 3.
Classic SAM segments one thing per prompt, which is ideal for interactive editing. Concept-promptable models take a phrase or an example and return masks for every match in the image, which is what labeling, counting, and scene analytics actually need. If your workflows currently chain a detector into SAM to get that behavior, a concept-promptable model folds it into one step.
The sweet spot is open-world and one-off jobs: labeling a new dataset, ad hoc analysis, editing pipelines. For a fixed, high-volume segmentation task, a compact model trained on masks the promptable model helped label still wins on cost and latency, the same trade as everywhere else on this site.
The bottom line: Reach for concept segmentation when you need every instance of something you can describe, and graduate fixed high-volume tasks to a trained model once the concept stops changing.
Promptable concept segmentation is the task of segmenting every instance of a concept you specify with a prompt, such as a text phrase, an example box, or a point, and returning instance-level masks for all matches. Where classic SAM segments "whatever is at this click", concept segmentation as in SAM 3 finds and masks every occurrence, merging open-vocabulary detection with segmentation in one pass: "every yellow cone" yields a mask per cone. This removes the need to train a custom segmentation model for many one-off jobs. It is used for interactive dataset labeling, targeted image editing, and open-world scene analytics. This page lists 6 promptable concept segmentation models, including 1 open-weight option you can self-host; all of them run live in the Playground so you can test them on your own images.
It depends on your task and constraints. For fixed categories in production, a model fine-tuned on your own data typically beats any general-purpose model. Compare the promptable concept segmentation models on this page and try them on your own images to see which fits.
Yes. 1 of the 6 promptable concept segmentation models here are open-weight (for example SAM 3), free to self-host under their licenses (Custom).
Yes. You can run all 6 in the Roboflow Playground for free. Upload an image and compare the models' output side by side, no setup required.
This page lists all 6 promptable concept segmentation models in the Roboflow Playground catalog: 1 open-weight model you can self-host and 5 proprietary models accessed through provider APIs; all of them run live in the Roboflow Playground on your own images. Compare licenses, parameters, API prices, and release dates side by side, or open any model page for full details.