What's new in ArcGIS AI models

New and updated ArcGIS AI models are described below along with links for additional information.

July 2026 release

New models

• Clay

  Wavelength-aware deep learning model to generate semantic geospatial embeddings from multisensor satellite imagery and enable efficient fine-tuning for downstream applications.

• DINOv3

  Deep learning model to generate semantic geospatial embeddings from RGB satellite imagery.

• Geospatial Vision Language Model (GeoVLM)

  Esri's first vision-language model for high-resolution RGB Earth observation imagery. This model will be available shortly as a private beta through Esri's Early Adopter Community. Please sign up here to request access to the model.

• USA Geodemographic Embeddings

  Esri's USA Geodemographic Embeddings 2026 include compact embeddings representing geodemographic and environmental characteristics across the United States. Use them to enrich models, find similar locations, and power geospatial AI workflows. (Beta)

June 2026 release

New models

• DINOv2

  Generate semantic geospatial embeddings from RGB satellite imagery.

• DistilGPT2 Based Imputation

  Impute missing numerical values in datasets using row-wise next token predictions.

• DOFA (Large)

  Wavelength-aware deep learning model to generate semantic geospatial embeddings from multisensor satellite imagery.

• Global Location Encoder (Sentinel-2)

  A representation learning model that maps geographic coordinates to meaningful vector embeddings based on Sentinel-2 Level-2.

• HF Text Feature Extraction

  Deep learning package for generating text embeddings using Hugging Face sentence-transformers models.

• Prithvi EO 2.0 300M

  Deep learning model to generate semantic geospatial embeddings from multispectral satellite imagery and enable efficient fine-tuning for downstream applications.

• Prithvi EO 2.0 600M

  Deep learning model to generate semantic geospatial embeddings from multispectral satellite imagery and enable efficient fine-tuning for downstream applications.

• Sen2SR Sentinel-2 Super-resolution

  Deep learning model to enhance the spatial resolution of Sentinel-2 imagery.

• TerraMind (Base)

  Deep learning model to generate semantic geospatial embeddings from multispectral satellite imagery.

• XGBoost Based Imputation

  Deep learning package using XGBoost for predictive imputation of missing numeric values in datasets.

September 2025 release

New models

• HF Entity Recognition

  Extract entities from text.

• HF Text Classification

  Classify text.

• HF Text Question Answering

  Perform question answering on a provided context text.

• HF Text Summarization

  Summarize text.

• HF Text Translation

  Translate text.

• HF Zero Shot Text Classification

  Classify text into user-defined categories without requiring model retraining.

June 2025 release

New models

• HF Image Inpainting

  Inpaint masked areas using pretrained image inpainting models.

• Prithvi Weather and Climate

  Forecast weather and climate related phenomenon using MERRA-2 multidimensional dataset.

• Table Extraction

  Extract tables from scanned images or PDFs.

• Time-MoE

  Forecast timeseries values of a space-time cube dataset using a pretrained model.

Updated models

• Clip Zero-Shot Classifier

  Added transformers version fix.

• Damage Assessment (Drone Imagery)

  Added multilabel Support.

Updated documentation

• Using Clip Zero-Shot Classifier

  Recommendation regarding fishnet grid generation for full image processing updated.

• Introduction to Damage Assessment (Drone Imagery)

  Model overview and recommendation regarding fishnet grid generation for full image processing updated.

• Using Damage Assessment (Drone Imagery)

  Steps for using the model updated.

• Finetuning the High Resolution Land Cover Classification - USA

  Lama based model inference is supported only on GPUs.

• Using Image Interrogation

  Lama based model inference is supported only on GPUs.

• Finetuning the Prithvi - Burn Scars Segmentation

  Steps to fine tune the model updated.

• Finetuning the Prithvi - Crop Classification

  Steps to fine tune the model updated.

• Finetuning the Prithvi - Flood Segmentation

  Steps to fine tune the model updated.

• Using Prithvi - Burn Scars Segmentation

  Recommended to choose GPU to achieve optimal performance.

• Using Prithvi - Crop Classification

  Recommended to choose GPU to achieve optimal performance.

• Using Prithvi - Flood Segmentation

  Recommended to choose GPU to achieve optimal performance.

• Using Vision Language Context-Based Classification

  Lama based model inference is supported only on GPUs.

March 2025 release

New model

• Image Interrogation

  Query images using prompts.

Updated models

• Ship Detection (RGB)

  Released version 2 of the model.

• Vision Language Context-Based Classification

  Released version 2 of the model.

Updated documentation

• Fine-tuning the Building Point Classification

  Steps to fine-tune the Building Point Classification pretrained model.

• Using Parking Lot Classification - USA

  Updated links.

• Fine-tuning the Power Line Classification

  Steps to fine-tune the Power Line Classification pretrained model.

• Introduction to Ship Detection (RGB)

  Updated with new information about the version 2 model.

• Fine-tuning the Tree Point Classification

  Steps to fine-tune the Tree Point Classification pretrained model.

• Introduction to Vision Language Context-Based Classification

  Updated with new information about the version 2 model.

• Using Vision Language Context-Based Classification

  Updated information about using Llama vision models.

February 2025 release

New models

• HF Object Classification

  Classify objects using Hugging Face object classification models.

• HF Object Detection

  Detect objects using Hugging Face object detection models.

• HF Pixel Classification

  Classify pixels using Hugging Face image segmentation models.

• HF Visual Question Answering

  Generate answers to open-ended questions, based on an image, using Hugging Face Visual Question Answering models.

• HF Zero Shot Object Classification

  Classify objects using Hugging Face zero-shot classification models.

Updated documentation

• Updated Fine-tuning steps for all models.
• Updated license information format for all models.

December 2024 release

New models

• Edge Detection

  Detect prominent edges present in an image.

• High Resolution Canopy Height Estimation

  Estimate tree canopy height in high-resolution satellite imagery.

• Map Simplification

  Simplifies scanned images to assist with digitization workflows.

• Pedestrian Infrastructure Classification

  Classify crosswalks, sidewalks, and roads from high resolution RGB imagery.

• Prompt Based Segmentation

  Segment objects using prompts from 8-bit RGB imagery.

• Vision Language Context-Based Classification

  Classify objects using OpenAI's vision language models from 8-bit RGB imagery.

Updated documentation

• Using Water Body Extraction (SAR) - USA

  Updated steps for model application.

October 2024 release

New models

• Building Change Detection

  Detect change in buildings in drone or aerial imagery.

• Damage Assessment (Drone Imagery)

  Perform damage assessment on drone and aerial imagery.

• Depth Anything

  Estimate relative and metric depth in street-level imagery.

Updated documentation

• Using Building Footprint Extraction - Africa

  Updated steps for model application using the Extract Feature tool.

• Using building footprint - Australia

  Updated steps for model application using the Extract Feature tool.

• Using building footprint - China

  Updated steps for model application using the Extract Feature tool.

• Using building footprint - USA

  Updated steps for model application using the Extract Feature tool.

• Using GroundingDINO

  Updated TTA parameter.

• Using Text SAM

  Updated TTA parameter.

• Updated ArcGIS Online license requirements information for all the existing pretrained models.

August 2024 release

New models

• GroundingDINO

  Detect objects using text prompts.

• Wildfire And Smoke Classification

  Classify wildfire and smoke in drone imagery.

Updated model

• Road Extraction - Global

  Updated Version 2 of the model. The previous version is now deprecated.

Updated documentation

• Oil Spill Detection (SAR)

  Data preparation steps updated.

• Road Extraction - North America

  Data preparation steps updated.

June 2024 release

New models

• Building Point Classification

  Classify point cloud into building or background.

• Clip Zero-Shot Classifier

  Classify oriented imagery.

• Wildfire Delineation

  Delineate wildfires using Sentinel-2 imagery.

Updated documentation

• Common Object Detection

  Model access and download detail updated.

• Face Blurring

  From ArcGIS Pro version 3.3 onward, this model will run on CPU, with GPU support discontinued.

• Fine-tuning the Ship Detection (RGB)

  Steps to fine-tune the Ship Detection (RGB) model.

• License Plate Blurring

  From ArcGIS Pro version 3.3 onward, this model will run on CPU, with GPU support discontinued.

• Mangrove classification (Landsat 8)

  Data preparation step updated.

• Named Entity Recognition

  Model details updated.

• Power Line Classification

  Model details updated.

• Road Extraction - Global

  From ArcGIS Pro version 3.3 onward, this model will run on CPU, with GPU support discontinued.

• Tree Point Classification

  Model details updated.

May 2024 release

Updated documentation

• Fine-tuning the Arctic Seal Detection

  Steps to fine-tune the Arctic Seal Detection pretrained model.

• Fine-tuning the Cloud Mask Generation (Sentinel-2)

  Steps to fine-tune the Cloud Mask Generation (Sentinel-2) pretrained model.

• Using Ship Detection (SAR)

  New steps utilizing the manifest.safe file as the input raster for inference.

• Using Water Body Extraction (SAR) - USA

  New steps utilizing the manifest.safe file as the input raster for inference.

April 2024 release

Updated documentation

• Fine-tuning Address Standardization

  Steps to fine-tune the Address Standardization pretrained model.

• Fine-tuning Agricultural Field Delineation

  Steps to fine-tune the Agricultural Field Delineation pretrained model.

• Fine-tuning Building Footprint Extraction - Australia

  Steps to fine-tune the Building Footprint Extraction - Australia pretrained model.

• Fine-tuning Building Footprint Extraction - China

  Steps to fine-tune the Building Footprint Extraction - China pretrained model.

• Fine-tuning Cooling Tower Detection - USA

  Steps to fine-tune Cooling Tower Detection - USA pretrained model.

• Fine-tuning Country Classification

  Steps to fine-tune the Country Classification pretrained model.

• Fine-tuning Elephant Detection

  Steps to fine-tune the Elephant Detection pretrained model.

• Fine-tuning Human Detection (Drone Imagery)

  Steps to fine-tune the Human Detection (Drone Imagery) pretrained model.

• Fine-tuning Human Settlements Classification (Landsat 8)

  Steps to fine-tune the Human Settlements Classification (Landsat 8) pretrained model.

• Fine-tuning Human Settlements Classification (Sentinel-2)

  Steps to fine-tune the Human Settlements Classification (Sentinel-2) pretrained model.

• Fine-tuning Insulator Defect Detection

  Steps to fine-tune the Insulator Defect Detection pretrained model.

• Fine-tuning Mangrove Clasification (Landsat 8)

  Steps to fine-tune the Mangrove Clasification (Landsat 8) pretrained model.

• Fine-tuning Oil Spill Detection (SAR)

  Steps to fine-tune the Oil Spill Detection (SAR) pretrained model.

• Fine-tuning Oil Tank Detection

  Steps to fine-tune the Oil Tank Detection pretrained model.

• Fine-tuning Palm Tree Detection

  Steps to fine-tune the Palm Tree Detection pretrained model.

• Fine-tuning Parcel Extraction - USA

  Steps to fine-tune the Parcel Extraction - USA pretrained model.

• Fine-tuning Parking Lot Classification

  Steps to fine-tune the Parking Lot Classification - USA pretrained model.

• Fine-tuning Parking Spot Detection - USA

  Steps to fine-tune the Parking Spot Detection - USA pretrained model.

• Fine-tuning Plant Leaf Disease Classification

  Steps to fine-tune the Plant Leaf Disease Classification pretrained model.

• Fine-tuning Prithvi - Crop Classification

  Steps to fine-tune the Prithvi - Crop Classification pretrained model.

• Fine-tuning Prithvi - Flood Segmentation

  Steps to fine-tune the Prithvi - Flood Segmentation pretrained model.

• Fine-tuning Pylon Detection - USA

  Steps to fine-tune the Pylon Detection - USA pretrained model.

• Fine-tuning Seabird (Tern) Detection - Africa

  Steps to fine-tune the Seabird (Tern) Detection - Africa pretrained model.

• Fine-tuning Ship Detection (SAR)

  Steps to fine-tune the Ship Detection (SAR) pretrained model.

• Fine-tuning Solar Panel Detection - USA

  Steps to fine-tune the Solar Panel Detection - USA pretrained model.

• Fine-tuning Solar Photovoltaic Park Classification

  Steps to fine-tune the Solar Photovoltaic Park Classification pretrained model.

• Fine-tuning Transmission H-Structure Detection

  Steps to fine-tune the Transmission H-Structure Detection pretrained model.

• Fine-tuning Well Pad Detection - Permian Basin

  Steps to fine-tune the Well Pad Detection - Permian Basin pretrained model.

• Fine-tuning Wind Turbine Detection

  Steps to fine-tune the Wind Turbine Detection pretrained model.

March 2024 release

New model

• Text SAM

  Detect objects in an image using a text prompt.

February 2024 release

Updated documentation

• Fine-tuning Building Footprint Extraction - Africa

  Steps to fine-tune the Building Footprint Extraction - Africa pretrained model.

• Fine-tuning Building Footprint Extraction - USA

  Steps to fine-tune the Building Footprint Extraction - USA pretrained model.

• Fine-tuning Car Detection - USA

  Steps to fine-tune the Car Detection - USA pretrained model.

• Fine-tuning High Resolution Land Cover Classification - USA

  Steps to fine-tune the High Resolution Land Cover Classification - USA pretrained model.

• Fine-tuning Land Cover Classification (Aerial Imagery)

  Steps to fine-tune the Land Cover Classification (Aerial Imagery) pretrained model.

• Fine-tuning Land Cover Classification (Landsat 8)

  Steps to fine-tune the Land Cover Classification (Landsat 8) pretrained model.

• Fine-tuning Land Cover Classification (Sentinel-2)

  Steps to fine-tune the Land Cover Classification (Sentinel-2) pretrained model.

• Fine-tuning Pool Detection - USA

  Steps to fine-tune the Pool Detection - USA pretrained model.

• Fine-tuning Prithvi - Burn Scars Segmentation

  Steps to fine-tune the Prithvi - Burn Scars Segmentation pretrained model.

• Fine-tuning Road Extraction - North America

  Steps to fine-tune the Road Extraction - North America pretrained model.

• Fine-tuning Waterbody Extraction (SAR) - USA

  Steps to fine-tune the Waterbody Extraction (SAR) - USA pretrained model.

January 2024 release

New models

• Agricultural Field Delineation

  Delineate agricultural fields using Sentinel-2 imagery.

• Prithvi - Burn Scars Segmentation

  Classify burn scars in multispectral satellite imagery.

• Prithvi - Crop Classification

  Classify crops in multispectral satellite imagery.

• Prithvi - Flood Segmentation

  Classify flood inundated areas in multispectral satellite imagery.

Updated models

• Cloud Mask Generation

  Radiometric offset correction parameter added.

• Human Settlements Classification (Sentinel-2)

  Radiometric offset correction parameter added.

• Land Cover Classification (Sentinel-2)

  Radiometric offset correction parameter added.

• Power Line Classification

  Update includes enhancements in architecture, increased precision, and faster processing.

• Solar Photovoltaic Park Classification

  Radiometric offset correction parameter added.

• Well Pad Detection - Permian Basin

  Radiometric offset correction parameter added.

Updated documentation

• Cloud Mask Generation

  Data preparation step updated.

• Human Settlements Classification (Sentinel-2)

  Data preparation step updated.

• Land Cover Classification (Sentinel-2)

  Data preparation step updated.

• Power Line Classification

  Updated information about the utilization of a new architecture, and revised accuracy metrics.

• Solar Photovoltaic Park Classification

  Data preparation step updated.

• Well Pad Detection - Permian Basin

  Data preparation step updated.

November 2023 release

New models

• Cooling Tower Detection—USA

  Detect cooling towers in high-resolution (4-15 centimeters) imagery.

• Oil Tank Detection

  Detect oil tanks in high-resolution satellite imagery.

• Pylon Detection—USA

  Detect pylons in high-resolution satellite imagery.

• Ship Detection (RGB)

  Detect ships in high-resolution satellite imagery.

Updated models

• Building Footprint Extraction—Australia

  SatToGraph issue fixed.

• Palm Tree Detection

  Undefined class name issue fixed.

• Road Extraction—Global

  Dlpk Issue fixed.

• Scene Text Parsing

  Pro3.2 issue fixed.

Updated documentation

• Arctic Seal Detection

  TTA (Test Time Augmentation) parameter details updated.

• Elephant Detection

  TTA (Test Time Augmentation) parameter details updated.

• Insulator Defect Detection

  TTA (Test Time Augmentation) parameter details updated.

• Pool Detection - USA

  TTA (Test Time Augmentation) parameter details updated.

• Transmission H-Structure Detection

  TTA (Test Time Augmentation) parameter details updated.

• Tree Detection

  Training data information updated.

• Scene Text Parsing

  PaddleOCR installation command updated.

• Wind Turbine Detection

  TTA (Test Time Augmentation) parameter details updated.

July 2023 release

New model

• Insulator Defect Detection

  Detect insulators and classify defects.

• Optical Character Recognition

  Detect and recognize text from images.

• Parking Lot Classification—USA

  Detect and classify pixels of parking areas from high-resolution aerial imagery.

• Pool Segmentation—USA

  Detect and segment swimming pools in high-resolution aerial or satellite imagery.

• Transmission H-Structure Detection

  Detect transmission H-Structure and its different parts.

• Tree Segmentation

  Detect and segment trees in high-resolution imagery.

Updated model

• Common Object Detection

  Detect common objects in context for mobile apps.

May 2023 release

New model

• Segment Anything Model (SAM)

  Segment any objects in any imagery.

April 2023 release

New model

• Parking Spot Detection—USA

  Detect parking spots from high resolution imagery.

Model Updates

• Building Footprint Extraction—Africa

  Added support for use with extract features tool.

• Building Footprint Extraction—Australia

  Added support for use with extract features tool.

• Building Footprint Extraction—USA

  Added support for use with extract features tool.

• Scene Text Parsing

  Now supported only with CPU.

February 2023 release

New model

• Common Object Detection

  Detect common objects in photos that are captured with the mobile app ArcGIS Survey123.

Updated model

• Human Detection (Drone Imagery)

  Detect human beings on drone imagery.

November 2022 release

New models

• Oil Spill Detection (SAR)

  Extract oil spill from SAR imagery.

• Plant Leaf Disease Classification

  Classify diseases in plant leaves.

September 2022 release

New models

• Building Footprint Extraction—China

  Extract building footprints from high-resolution three-band imagery for China.

• Seabird (Tern) Detection—Africa

  Detect seabirds (Tern) from high-resolution drone imagery.

• Land Cover Classification (Aerial Imagery)

  Perform land-cover classification on very high-resolution imagery.

• Water Body Extraction (SAR)—USA

  Extract water bodies from SAR imagery.

July 2022 release

New models

• Address Standardization

  Transforms non-standard address to standard address.

• Cloud Mask Generation (Sentinel-2)

  Detect and generate cloud mask from satellite imagery.

• Country Classification

  Classify addresses into their respective countries.

May 2022 release

New models

• Elephant Detection

  Detect elephants from high-resolution drone imagery.

• Named Entity Recognition

  Identify entities from text.

• Solar Photovoltaic Park Classification

  Detect solar photovoltaic parks from Sentinel-2 imagery.

• Tree Detection

  Detect trees in high-resolution drone or aerial imagery.

March 2022 release

New models

• Arctic Seal Detection

  Detect arctic seals from high-resolution drone imagery.

• Pavement Crack Detection

  Detect pavement cracks and potholes in drone imagery.

• Scene Text Parsing

  Recognize text from a variety of image formats and extract information in the form of text from images.

• Wind Turbine Detection

  Detect wind turbines from high-resolution imagery.

Updated models

• Car Detection—USA

  Added support for test time augmentation and regularization of results.

• High Resolution Land Cover Classification—USA

  Added support for test time augmentation and classification of images in seven or nine classes.

• Human Settlements Classification (Landsat 8)

  Added support for test time augmentation.

• Human Settlements Classification (Sentinel-2)

  Added support for test time augmentation.

• Land Cover Classification (Landsat 8)

  Added support for test time augmentation and bundled two different models to support classification on level-1 and level-2 images.

• Land Cover Classification (Sentinel-2)

  Added support for test time augmentation and bundled four different models supporting classification of Sentinel-2 Level-2A and Sentinel-2 Level-1C into CLC level 1 and level 2 classes.

• License Plate Blurring

  Added support for blurring faces in the images.

• Mangrove Classification (Landsat 8)

  Added support for test time augmentation.

• Pool Detection—USA

  Added support for test time augmentation.

January 2022 release

New models

• Building Footprint Extraction—Australia

  Extract building footprints from high-resolution three-band imagery for Australia.

• High Resolution Land Cover Classification—USA

  Perform land-cover classification on high-resolution imagery.

• Human Detection (Drone Imagery)

  Detect human beings on drone imagery.

• Palm Tree Detection

  Detect palm trees on high-resolution imagery.

• Road Extraction—Global

  Extract roads from high-resolution (1 meter) aerial/satellite imagery.

Updated models

• Road Extraction—North America

  Added support for test time augmentation and smooth blending of results.

• Land Cover Classification (Landsat 8)

  Model is now compatible with Landsat 8 Collection 2 Level 2 images.

November 2021 release

New models

• Mangrove Classification (Landsat 8)

  Classify mangroves from Landsat 8 imagery.

• Well Pad Detection—Permian Basin

  Detect well pads on Sentinel-2 imagery.

Updated models

• Power Line Classification

  Made improvements to precision and recall, and simplified class structure.

July 2021 release

New models

• Building Footprint Extraction—Africa

  Extract building footprints from high-resolution three-band imagery for Africa.

• Parcel Extraction—USA

  Demarcate land parcel boundaries with plausible approximations for the United States.

• Object Tracking

  Use for applications such as object follower and surveillance of stationary objects.

• Pool Detection—USA

  Detect pools from high-resolution three-band imagery for the United States.

• Solar Panel Detection—USA

  Detect solar panels from high-resolution three-band imagery for the United States.

• Crowd Counting

  Count crowds.

• Car Detection—USA

  Detect cars from high-resolution three-band imagery for the United States.

• Ship Detection (SAR)

  Detect ships globally from SAR data.

February 2021 release

New models

• Road Extraction—North America

  Extract roads from high-resolution (30–50 centimeters) aerial and satellite imagery.

• Land Cover Classification (Sentinel-2)

  Perform land cover classification on Sentinel-2 imagery.

• Human Settlements Classification (Sentinel-2)

  Classify human settlements on Sentinel-2 imagery.

• Human Settlements Classification (Landsat 8)

  Classify human settlements on Landsat 8 imagery.

• Shipwreck Detection

  Detect shipwrecks using BAG data.

• License Plate Blurring

  Blur car license plates to maintain privacy. As democratization, availability, and acquisition of data is increasing, privacy is becoming a concern. Regulations such as GDPR have been introduced to address the privacy rights of the users.

• Face Blurring

  Blur faces to maintain privacy. As democratization, availability, and acquisition of data is increasing, privacy is becoming a concern. Regulations such as GDPR have been introduced to address the privacy rights of the users.

• Windows and Doors Extraction

  Extract windows and doors in textured building data displayed in 3D views.

Updated models

• Building Footprint Extraction—USA

  Extract building footprints from high-resolution three-band imagery for the United States.

December 2020 release

New models

• Power Line Classification

  Extract distribution wires at the street level from point cloud data.

October 2020 release

New models

• Building Footprint Extraction—USA

  Extract building footprints from high-resolution three-band imagery for the United States.

• Land Cover Classification (Landsat 8)

  Classify land cover on Landsat 8 imagery.

• Tree Point Classification

  Classify trees from point cloud data.