MetaSpecialist modelLimited access
SAM 3
75out of 100new calculation
based on two data groups, mostly search, range 66 to 85
Summary
SAM 3 is Meta’s foundation model for promptable concept segmentation in images and videos. Text, points, boxes, masks, and image exemplars can describe objects. The model detects, segments, and tracks instances instead of generating new images.
The Hugging Face checkpoint is gated and requires consent to share contact information. The weights and official code are intended for local use.
On this page
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 310 published benchmark results
Page 1 of 26
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
CountBench accuracy (Show measurement, test conditions, and source)
- Source value
- 93.8
- Metric
- percent
- Unit
- %
- Category
- object-counting
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SAM 3 technical paper
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 3 object counting.
PixMo-Count accuracy (Show measurement, test conditions, and source)
- Source value
- 86.2
- Metric
- percent
- Unit
- %
- Category
- object-counting
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SAM 3 technical paper
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 3 object counting.
SA-37 mean IoU, one click (Show measurement, test conditions, and source)
- Source value
- 66.1
- Metric
- percent
- Unit
- %
- Category
- segmentation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SAM 3 technical paper
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 6 one-click interactive segmentation.
SA-37 mean IoU, three clicks (Show measurement, test conditions, and source)
- Source value
- 81.3
- Metric
- percent
- Unit
- %
- Category
- segmentation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SAM 3 technical paper
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 6 three-click interactive segmentation.
SA-37 mean IoU, five clicks (Show measurement, test conditions, and source)
- Source value
- 85.1
- Metric
- percent
- Unit
- %
- Category
- segmentation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SAM 3 technical paper
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Table 6 five-click interactive segmentation.
ODinW13 bbox mAP, 0-shot transfer (Show measurement, test conditions, and source)
- Source value
- 61
- Metric
- average bbox mAP, AP0
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Meta, SAM 3 technical paper, version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. SAM 3 checkpoint; Table 2 zero-shot transfer; original benchmark labels as prompts; average bbox mAP.
- Context
- The paper does not report a scored sample count or confidence interval for this result.
ODinW13 bbox mAP, 10-shot transfer (Show measurement, test conditions, and source)
- Source value
- 71.8
- Metric
- average bbox mAP, AP10
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Meta, SAM 3 technical paper, version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. SAM 3 checkpoint fine-tuned on 10 examples per benchmark class without mask loss; no prompt tuning; average bbox mAP.
- Context
- This is a 10-shot adapted result, not an untouched checkpoint result. The paper does not report a scored sample count or confidence interval.
RF100VL bbox mAP, 0-shot transfer (Show measurement, test conditions, and source)
- Source value
- 15.2
- Metric
- average bbox mAP, AP0
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Meta, SAM 3 technical paper, version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. SAM 3 checkpoint; Table 2 zero-shot transfer; original benchmark labels as prompts; average bbox mAP.
- Context
- The paper does not report a scored sample count or confidence interval for this result.
RF100VL bbox mAP, 10-shot transfer (Show measurement, test conditions, and source)
- Source value
- 36.5
- Metric
- average bbox mAP, AP10
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Meta, SAM 3 technical paper, version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. SAM 3 checkpoint fine-tuned on 10 examples per benchmark class without mask loss; no prompt tuning; average bbox mAP.
- Context
- This is a 10-shot adapted result, not an untouched checkpoint result. The paper does not report a scored sample count or confidence interval.
SA-Co/VEval SA-V cgF1 (Show measurement, test conditions, and source)
- Source value
- 30.3
- Metric
- classification-gated positive micro F1
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Meta, SAM 3 technical paper, version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. SAM 3 video PCS; text noun-phrase prompt; SA-Co/VEval SA-V subset; Table 4.
- Context
- The reported NPs are prompt counts in the table heading, not the scored sample denominator. No denominator or confidence interval is provided.
SA-Co/VEval SA-V pHOTA (Show measurement, test conditions, and source)
- Source value
- 58
- Metric
- pHOTA
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Meta, SAM 3 technical paper, version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. SAM 3 video PCS; text noun-phrase prompt; SA-Co/VEval SA-V subset; Table 4.
- Context
- The reported NPs are prompt counts in the table heading, not the scored sample denominator. No denominator or confidence interval is provided.
SA-Co/VEval YT-Temporal-1B cgF1 (Show measurement, test conditions, and source)
- Source value
- 50.8
- Metric
- classification-gated positive micro F1
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Meta, SAM 3 technical paper, version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. SAM 3 video PCS; text noun-phrase prompt; SA-Co/VEval YT-Temporal-1B subset; Table 4.
- Context
- The reported NPs are prompt counts in the table heading, not the scored sample denominator. No denominator or confidence interval is provided.
Page 1 of 26
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 21 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 20, 2026
SAM 3EmbeddingGemma 2 70Unlimited OCR 2 55Kumo-Tabular 45TeleOCR 38
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
- 2November 1, 2026: Method version 4 (common scale, shrinkage)
- 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as SAM 3, chosen by model line, developer, size, price and release date.
- MetaEncoder v1Meta · same developer
- EmbeddingGemma 2Google DeepMind · popular in this category
- Unlimited OCR 2Baidu · popular in this category
- Kumo-TabularNVIDIA · popular in this category
- TeleOCRTeleAI · popular in this category
- Qwen-Drive-1.0-4BQwen · popular in this category
- AstaBrief-8BAllen Institute for AI · popular in this category
- Agora-2Odyssey · popular in this category
More about SAM 3
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
- Shutdown
- None announced
- no source
Further details
Sources and data date
Show all 15 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.