{"slug": "segment-anything-3", "name": "Segment Anything 3", "category": "ai-models", "category_name": "AI models", "description": "", "recommenders": 13, "commenters": 21, "warners": 1, "score": 10.943, "rising": -4.25, "first": "2025-11-19", "latest": "2026-08-09", "why": "When I swapped SAM 2 for SAM 3 the whole system improved", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 12, "neutral": 7, "compare": 1, "warn": 1}, "spark": [8, 1, 0, 1, 0, 0, 0, 1, 1, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "When I swapped SAM 2 for SAM 3 the whole system improved", "kind": "", "icon": true, "domain": null, "github_repo": "github.com/facebookresearch/sam3", "official_url": "https://ai.meta.com/sam3/", "pricing": "", "aliases": ["SAM 3", "SAM3", "Segment Anything", "Segment Anything 3", "Segment Anything Model 3"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 782505201, "item_id": 48906575, "url": "https://news.ycombinator.com/item?id=48906575", "by": "lapurita", "date": "2026-07-14", "version": "3", "stance": "recommend", "quote": "When I swapped SAM 2 for SAM 3 the whole system improved", "excerpt": "This started when I saw a Roboflow tutorial tracking basketball players by giving every player's box to SAM2 and propagating masks through the whole clip (https://www.youtube.com/watch?v=yGQb9KkvQ1Q). Identity preservation worked way better than any dedicated tracker I'd used before but it only ran at 1\u20132 fps, where both speed and memory was scaling with player count. Most frames don't need SAM at all though. A norma", "context": "", "use_case": "", "switched_from": "SAM 2", "story": null}, {"id": 775326160, "item_id": 48457885, "url": "https://news.ycombinator.com/item?id=48457885", "by": "fnands", "date": "2026-06-09", "version": "3", "stance": "recommend", "quote": "It's already pretty much possible with open-vocabulary detectors like SAM3", "excerpt": "That seems to be the way things are going. Large general models have taken over in NLP, and (outside of embedded/low latency applications) it seems like they are coming for CV next. So you should soon be able to have large generic model that can detect whatever for you. It's already pretty much possible with open-vocabulary detectors like SAM3, where you could just prompt it with \"Apple\": https://ai.meta.com/research", "context": "", "use_case": "", "switched_from": "", "story": "OpenCV 5 Is Here: The Biggest Leap in Years for Computer Vision"}, {"id": 750553888, "item_id": 46909618, "url": "https://news.ycombinator.com/item?id=46909618", "by": "joriJordan", "date": "2026-02-05", "version": "3", "stance": "recommend", "quote": "IMO that's far more valuable to the ecosystem than more AIaaS", "excerpt": "Meta research released updated segmentation models a few months ago: https://github.com/facebookresearch/sam-3d-objects https://github.com/facebookresearch/sam3 IMO that's far more valuable to the ecosystem than more AIaaS", "context": "", "use_case": "image segmentation models", "switched_from": "", "story": "10 months since the Llama-4 release: what happened to Meta AI?"}, {"id": 741828560, "item_id": 46364285, "url": "https://news.ycombinator.com/item?id=46364285", "by": "AkshatJ27", "date": "2025-12-23", "version": null, "stance": "recommend", "quote": "You can try it out in the playground", "excerpt": "You can try it out in the playground: https://aidemos.meta.com/segment-anything/gallery/ There seem to be many more fun little demos by meta here like automatic video masking, making 3d models from 2d images, etc.", "context": "", "use_case": "", "switched_from": "", "story": "Our New Sam Audio Model Transforms Audio Editing"}, {"id": 737116160, "item_id": 46069760, "url": "https://news.ycombinator.com/item?id=46069760", "by": "Glemkloksdjf", "date": "2025-11-27", "version": null, "stance": "recommend", "quote": "Segment Anything? Yepp we use it.", "excerpt": "I'm still ahead of a company starting later: I have the legal structure, i know my collegues, i have potentially employees and more capacity. The problem is not that a startup is starting after you but you do not give yourself time to keep an eye on AI and not leveraging it when its helpful. We leverage AI and ML progress constantly and keep an eye on advances. Segment Anything? Yepp we use it. Claude? Yes Sir!", "context": "runs a company", "use_case": "", "switched_from": "", "story": "Technical Deflation"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 783951888, "item_id": 48996993, "url": "https://news.ycombinator.com/item?id=48996993", "by": "commoner", "date": "2026-07-21", "version": "3", "stance": "warn", "quote": "Meta's models are not open source because their weights always come with weird restrictions.", "excerpt": "Whenever Meta claims their models are open source, you have to double-check. SAM License (https://github.com/facebookresearch/sam3/blob/main/LICENSE): > iv. Your use of the SAM Materials will not involve or encourage others to reverse engineer, decompile or discover the underlying components of the SAM Materials. > v. You are not the target of Trade Controls and your use of SAM Materials must comply with Trade Contro", "context": "", "use_case": "", "switched_from": "", "story": "Meta's AI Models Are Powering the First Wave of Genesis Mission Projects"}]}], "who": [{"by": "Glemkloksdjf", "context": "runs a company", "url": "https://news.ycombinator.com/item?id=46069760", "date": "2025-11-27"}, {"by": "tom-in-july", "context": "medical imaging researcher", "url": "https://news.ycombinator.com/item?id=45986666", "date": "2025-11-19"}], "tags": [], "use_cases": ["image segmentation", "image masking", "skier tracking and pose estimation", "image segmentation models", "text-prompted segmentation", "video segmentation with streaming"], "compared": [{"slug": "orion", "name": "Orion", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "dinov3", "name": "DINOv3", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [{"id": 735702400, "item_id": 45981400, "url": "https://news.ycombinator.com/item?id=45981400", "by": "yeldarb", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "It's really, really good.", "excerpt": "We (Roboflow) have had early access to this model for the past few weeks. It's really, really good. This feels like a seminal moment for computer vision. I think there's a real possibility this launch goes down in history as \"the GPT Moment\" for vision. The two areas I think this model is going to be transformative in the immediate term are for rapid prototyping and distillation. Two years ago we released autodistill", "context": "", "use_case": "Image segmentation model", "switched_from": "", "story": "Segment Anything 3"}], "threads": [{"url": "https://news.ycombinator.com/item?id=45982073", "title": "Meta Segment Anything Model 3", "mentions": 11}, {"url": "https://news.ycombinator.com/item?id=45981514", "title": "Meta Segment Anything Model 3", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=45981155", "title": "Segment Anything 3", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49233423", "title": "Ask HN: What are you working on? (August 2026)", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48995074", "title": "Meta's AI Models Are Powering the First Wave of Genesis Mission Projects", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48906574", "title": "item 48906574", "mentions": 1}], "sentiment_by_month": {"2025-11": {"recommend": 8, "neutral": 3, "compare": 1}, "2025-12": {"neutral": 1, "recommend": 1}, "2026-01": {}, "2026-02": {"recommend": 1}, "2026-03": {"neutral": 1}, "2026-04": {"neutral": 1}, "2026-05": {}, "2026-06": {"recommend": 1}, "2026-07": {"recommend": 1, "warn": 1}, "2026-08": {"neutral": 1}, "2026-09": {}, "2026-10": {}}, "versions": [], "version_unnamed": 3, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 787747793, "item_id": 49234237, "url": "https://news.ycombinator.com/item?id=49234237", "by": "karolist", "date": "2026-08-09", "version": null, "stance": "neutral", "quote": "SAM-assisted selection", "excerpt": "I\u2019m working on PhotoFig: https://photofig.com, a recently launched product-photo editor for online sellers who need to clean up hundreds of images. It was inspired by my wife having to carefully mask product photos for her online shop. Existing tools could remove backgrounds automatically, but didn\u2019t give her enough control when the result needed correcting. So masking is a first-class feature, with SAM-assisted sele", "context": "", "use_case": "image masking", "switched_from": "", "story": "Ask HN: What are you working on? (August 2026)"}, {"id": 783951888, "item_id": 48996993, "url": "https://news.ycombinator.com/item?id=48996993", "by": "commoner", "date": "2026-07-21", "version": "3", "stance": "warn", "quote": "Meta's models are not open source because their weights always come with weird restrictions.", "excerpt": "Whenever Meta claims their models are open source, you have to double-check. SAM License (https://github.com/facebookresearch/sam3/blob/main/LICENSE): > iv. Your use of the SAM Materials will not involve or encourage others to reverse engineer, decompile or discover the underlying components of the SAM Materials. > v. You are not the target of Trade Controls and your use of SAM Materials must comply with Trade Contro", "context": "", "use_case": "", "switched_from": "", "story": "Meta's AI Models Are Powering the First Wave of Genesis Mission Projects"}, {"id": 782505201, "item_id": 48906575, "url": "https://news.ycombinator.com/item?id=48906575", "by": "lapurita", "date": "2026-07-14", "version": "3", "stance": "recommend", "quote": "When I swapped SAM 2 for SAM 3 the whole system improved", "excerpt": "This started when I saw a Roboflow tutorial tracking basketball players by giving every player's box to SAM2 and propagating masks through the whole clip (https://www.youtube.com/watch?v=yGQb9KkvQ1Q). Identity preservation worked way better than any dedicated tracker I'd used before but it only ran at 1\u20132 fps, where both speed and memory was scaling with player count. Most frames don't need SAM at all though. A norma", "context": "", "use_case": "", "switched_from": "SAM 2", "story": null}, {"id": 775326160, "item_id": 48457885, "url": "https://news.ycombinator.com/item?id=48457885", "by": "fnands", "date": "2026-06-09", "version": "3", "stance": "recommend", "quote": "It's already pretty much possible with open-vocabulary detectors like SAM3", "excerpt": "That seems to be the way things are going. Large general models have taken over in NLP, and (outside of embedded/low latency applications) it seems like they are coming for CV next. So you should soon be able to have large generic model that can detect whatever for you. It's already pretty much possible with open-vocabulary detectors like SAM3, where you could just prompt it with \"Apple\": https://ai.meta.com/research", "context": "", "use_case": "", "switched_from": "", "story": "OpenCV 5 Is Here: The Biggest Leap in Years for Computer Vision"}, {"id": 764123329, "item_id": 47757708, "url": "https://news.ycombinator.com/item?id=47757708", "by": "christianh", "date": "2026-04-13", "version": "3", "stance": "neutral", "quote": "", "excerpt": "I\u2019m working on https://poser.pro, a ski technique feedback tool for skiers that want to improve their skills. How? The user (skier!) uploads a short ski clip to Poser, lets Poser analyze, and then gets their results. For now, you get video outputs: head-tracking and skeleton overlay. I\u2019ve built a a lot of extra stuff (animated 3D body, turn-detection, balance-, steering-, pressure-, and edging-metrics, etc.), I\u2019m jus", "context": "", "use_case": "Skier tracking and pose estimation", "switched_from": "", "story": "Ask HN: What Are You Working On? (April 2026)"}, {"id": 756955537, "item_id": 47309721, "url": "https://news.ycombinator.com/item?id=47309721", "by": "subzel0", "date": "2026-03-09", "version": "3", "stance": "neutral", "quote": "it uses SAM3 to mask that region", "excerpt": "I\u2019m working on uvmap.ai A browser tool for editing 3D model textures directly from the model view instead of bouncing between a UV map, an image editor, and an AI tool. You load a glTF, click the part you want to change, it uses SAM3 to mask that region, then sends it to Nano Banana and puts the result back onto the texture. Still early, but the goal is to make texture iteration much less tedious. Soon here: https://", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: What Are You Working On? (March 2026)"}, {"id": 750553888, "item_id": 46909618, "url": "https://news.ycombinator.com/item?id=46909618", "by": "joriJordan", "date": "2026-02-05", "version": "3", "stance": "recommend", "quote": "IMO that's far more valuable to the ecosystem than more AIaaS", "excerpt": "Meta research released updated segmentation models a few months ago: https://github.com/facebookresearch/sam-3d-objects https://github.com/facebookresearch/sam3 IMO that's far more valuable to the ecosystem than more AIaaS", "context": "", "use_case": "image segmentation models", "switched_from": "", "story": "10 months since the Llama-4 release: what happened to Meta AI?"}, {"id": 741828560, "item_id": 46364285, "url": "https://news.ycombinator.com/item?id=46364285", "by": "AkshatJ27", "date": "2025-12-23", "version": null, "stance": "recommend", "quote": "You can try it out in the playground", "excerpt": "You can try it out in the playground: https://aidemos.meta.com/segment-anything/gallery/ There seem to be many more fun little demos by meta here like automatic video masking, making 3d models from 2d images, etc.", "context": "", "use_case": "", "switched_from": "", "story": "Our New Sam Audio Model Transforms Audio Editing"}, {"id": 738162496, "item_id": 46135156, "url": "https://news.ycombinator.com/item?id=46135156", "by": "iib", "date": "2025-12-03", "version": "3", "stance": "neutral", "quote": "I have been looking for the same thing, either from Meta's SAM 3[1] model", "excerpt": "I have been looking for the same thing, either from Meta's SAM 3[1] model, either from things like the OP. There has been some research specifically in this area with what appears to be classic ML models [2], but it's unclear to me if it can generalize to dances it has not been trained on. [1] https://ai.meta.com/blog/segment-anything-model-3/ [2] https://arxiv.org/html/2405.19727v1", "context": "", "use_case": "", "switched_from": "", "story": "Qwen3-VL can scan two-hour videos and pinpoint nearly every detail"}, {"id": 737116160, "item_id": 46069760, "url": "https://news.ycombinator.com/item?id=46069760", "by": "Glemkloksdjf", "date": "2025-11-27", "version": null, "stance": "recommend", "quote": "Segment Anything? Yepp we use it.", "excerpt": "I'm still ahead of a company starting later: I have the legal structure, i know my collegues, i have potentially employees and more capacity. The problem is not that a startup is starting after you but you do not give yourself time to keep an eye on AI and not leveraging it when its helpful. We leverage AI and ML progress constantly and keep an eye on advances. Segment Anything? Yepp we use it. Claude? Yes Sir!", "context": "runs a company", "use_case": "", "switched_from": "", "story": "Technical Deflation"}, {"id": 736774960, "item_id": 46048435, "url": "https://news.ycombinator.com/item?id=46048435", "by": "visioninmyblood", "date": "2025-11-25", "version": "3", "stance": "neutral", "quote": "you can download them at https://github.com/facebookresearch/sam3.", "excerpt": "you can download them at https://github.com/facebookresearch/sam3. for 3d https://github.com/facebookresearch/sam-3d-objects", "context": "", "use_case": "image segmentation", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 736764257, "item_id": 46047766, "url": "https://news.ycombinator.com/item?id=46047766", "by": "bradyriddle", "date": "2025-11-25", "version": "3", "stance": "recommend", "quote": "Sam 3 is incredible.", "excerpt": "Same. Checkout https://github.com/MiscellaneousStuff/meta-sam-demo It's a rip of the previous sam playground. I use it for a bunch of things. Sam 3 is incredible. I'm surprised it's not getting more attention.", "context": "", "use_case": "image segmentation", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735808289, "item_id": 45988018, "url": "https://news.ycombinator.com/item?id=45988018", "by": "featureofone", "date": "2025-11-19", "version": "3", "stance": "neutral", "quote": "the new text prompt input seams nice", "excerpt": "The SAM models are great. I used the latest version when building VideoVanish ( https://github.com/calledit/VideoVanish ) a video-editing GUI for removing or making objects vanish from videos. That used SAM 2, and in my experience SAM 2 was more or less perfect\u2014I didn\u2019t really see the need for a SAM 3. Maybe it could have been better at segmenting without input. But the new text prompt input seams nice; much easier t", "context": "", "use_case": "text-prompted segmentation", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735786656, "item_id": 45986666, "url": "https://news.ycombinator.com/item?id=45986666", "by": "tom-in-july", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "The native support for streaming in SAM3 is awesome.", "excerpt": "The native support for streaming in SAM3 is awesome. Especially since it should also remove some of the memory accumulation for long sequences. I used SAM2 for tracking tumors in real-time MRI images. With the default SAM2 and loading images from the da, we could only process videos with 10^2 - 10^3 frames before running out of memory. By developing/adapting a custom version (1) based on a modified implementation wit", "context": "medical imaging researcher", "use_case": "video segmentation with streaming", "switched_from": "SAM2", "story": "Meta Segment Anything Model 3"}, {"id": 735786224, "item_id": 45986639, "url": "https://news.ycombinator.com/item?id=45986639", "by": "nsingh2", "date": "2025-11-19", "version": "3", "stance": "neutral", "quote": "I'm not sure if the work they did with DINOv3 went into SAM3", "excerpt": "DINOv3 was released earlier this year: https://ai.meta.com/dinov3/ I'm not sure if the work they did with DINOv3 went into SAM3. I don't see any mention of it in the paper, though I just skimmed it.", "context": "", "use_case": "", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735785520, "item_id": 45986595, "url": "https://news.ycombinator.com/item?id=45986595", "by": "luckyLooking", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "Even works with long range shots.", "excerpt": "Even works with long range shots. https://chat.vlm.run/chat/e8bd5a29-a789-40aa-ae31-a510dc647886", "context": "", "use_case": "Image segmentation", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735783296, "item_id": 45986456, "url": "https://news.ycombinator.com/item?id=45986456", "by": "sorenjan", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "SAM3 is probably a great model to distill from", "excerpt": "SAM3 is probably a great model to distill from when training smaller segmentation models, but isn't their DINOv2 a better example of a large foundation model to distill from for various computer vision tasks? I've seen it used for as starting point for models doing segmentation and depth estimation. Maybe there's a v3 coming soon? https://dinov2.metademolab.com/", "context": "", "use_case": "", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735781840, "item_id": 45986365, "url": "https://news.ycombinator.com/item?id=45986365", "by": "visioninmyblood", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "you can download it at https://github.com/facebookresearch/sam3", "excerpt": "you can download it at https://github.com/facebookresearch/sam3. for 3d https://github.com/facebookresearch/sam-3d-objects I actually found the easiest way was to run it for free to see if it works for my use case of person deidentification https://chat.vlm.run/chat/63953adb-a89a-4c85-ae8f-2d501d30a429", "context": "", "use_case": "", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735779248, "item_id": 45986203, "url": "https://news.ycombinator.com/item?id=45986203", "by": "Benjamin_Dobell", "date": "2025-11-19", "version": "3", "stance": "compare", "quote": "SAM3 seems to less precisely trace the images", "excerpt": "For background removal (at least my niche use case of background removal of kids drawings \u2014 https://breaka.club/blog/why-were-building-clubs-for-kids) I think birefnet v2 is still working slightly better. SAM3 seems to less precisely trace the images \u2014 it'll discard kids drawing out the lines a bit, which is okay, but then it also seems to struggle around sharp corners and includes a bit of the white page that I'd li", "context": "", "use_case": "", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735778224, "item_id": 45986139, "url": "https://news.ycombinator.com/item?id=45986139", "by": "rocauc", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "I'm bullish SAM 3 can automate at least 90% of the work", "excerpt": "As someone that works on a platform users have used for labeling 1B images, I'm bullish SAM 3 can automate at least 90% of the work. Data prep is flipped to models being human-assisted instead of humans being model-assisted (see \"autolabel\" https://blog.roboflow.com/sam3/). I'm optimistic majority of users can now start deploying a model to then curate data instead of the inverse.", "context": "", "use_case": "", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735765345, "item_id": 45985334, "url": "https://news.ycombinator.com/item?id=45985334", "by": "rocauc", "date": "2025-11-19", "version": "3", "stance": "neutral", "quote": "SAM 3 - Open vocab concept segmentation on images and video", "excerpt": "A brief history. SAM 1 - Visual prompt to create pixel-perfect masks in an image. No video. No class names. No open vocabulary. SAM 2 - Visual prompting for tracking on images and video. No open vocab. SAM 3 - Open vocab concept segmentation on images and video. Roboflow has been long on zero / few shot concept segmentation. We've opened up a research preview exploring a SAM 3 native direction for creating your own m", "context": "", "use_case": "", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735756752, "item_id": 45984797, "url": "https://news.ycombinator.com/item?id=45984797", "by": "yeldarb", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "It's really, really good.", "excerpt": "We (Roboflow) have had early access to this model for the past few weeks. It's really, really good. This feels like a seminal moment for computer vision. I think there's a real possibility this launch goes down in history as \"the GPT Moment\" for vision. The two areas I think this model is going to be transformative in the immediate term are for rapid prototyping and distillation. Two years ago we released autodistill", "context": "", "use_case": "", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735756304, "item_id": 45984769, "url": "https://news.ycombinator.com/item?id=45984769", "by": "fzysingularity", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "SAM3 is cool", "excerpt": "SAM3 is cool - you can already do this more interactively on chat.vlm.run [1], and do much more. It's built on our new Orion [2] model; we've been able to integrate with SAM and several other computer-vision models in a truly composable manner. Video segmentation and tracking is also coming soon! [1] https://chat.vlm.run [2] https://vlm.run/orion", "context": "", "use_case": "image and video segmentation", "switched_from": "", "story": "Meta Segment Anything Model 3"}, {"id": 735751472, "item_id": 45984467, "url": "https://news.ycombinator.com/item?id=45984467", "by": "fzysingularity", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "SAM3 is cool", "excerpt": "SAM3 is cool - you can already do this more interactively on chat.vlm.run [1], and do much more. It's built on our new Orion [2] model; we've been able to integrate with SAM and several other computer-vision models in a truly composable manner. Video segmentation and tracking is also coming soon! [1] https://chat.vlm.run [2] https://vlm.run/orion", "context": "", "use_case": "", "switched_from": "", "story": "Segment Anything 3"}, {"id": 735744336, "item_id": 45984021, "url": "https://news.ycombinator.com/item?id=45984021", "by": "BoorishBears", "date": "2025-11-19", "version": "3", "stance": "recommend", "quote": "I've personally been waiting for SAM 3", "excerpt": "I've personally been waiting for SAM 3 which is also released now: https://github.com/facebookresearch/sam3", "context": "", "use_case": "", "switched_from": "", "story": null}]}