# Roboflow Playground

> A platform for discovering, comparing, and testing vision-language models (VLMs). It features a Model Directory for browsing 119+ models, a Playground for testing, an Arena for blind model comparisons, Rankings for performance leaderboards, and Vision Evals for ground-truth benchmark scoring across six real-world vision tasks. Supported tasks include Promptable Concept Segmentation (PCS), OCR, image captioning, object detection, multi-class classification, and visual question answering (VQA).

Roboflow Playground accepts image and text input, runs many models against the same input, and returns their outputs side by side. No login is required and every mode supports shareable result links. The site is maintained by Roboflow and lives at https://playground.roboflow.com.

## Capabilities

- Model Directory: browse 119+ vision models filtered by organization, task support, modality (vision, vision-language, multimodal), and access type (open or closed source).
- Individual model pages with detailed specs, performance metrics, ELO rankings, and usage statistics.
- Organization pages showing all models from providers like Google, Anthropic, OpenAI, Meta, and others.
- Four modes: Playground, Arena (blind comparisons), Rankings (leaderboards), and Vision Evals (ground-truth benchmark scoring across six real-world vision tasks).
- Accepts image plus text input (multimodal) and returns model outputs side by side for comparison.
- Anonymous voting system feeding an ELO-based ranking that tracks model performance over time.
- Vision Evals: ground-truth benchmark scoring of current vision-language models across six tasks (object detection, counting, identification, OCR, data extraction, reasoning), with measured token usage, cost, and speed per sample. Every model runs the same samples, with no human votes and no subjective judgment.
- Vision Evals per-model breakdowns on individual model pages, plus side-by-side eval results on compare pages. The previous version of Vision Evals (pass/fail scoring of 50+ models, including deprecated ones) is preserved at /evals/legacy.
- Deploy with API: Roboflow-provided models can be deployed via Roboflow Workflows. Fork a workflow to your workspace and get ready-to-use code snippets in Python, JavaScript, and cURL. Eligible model pages include an on-page Deploy section with the endpoint contract, code examples, and a ready-to-paste prompt for AI coding agents (see Deploying Models below).

## Modes

- [Model Directory](https://playground.roboflow.com/models): Browse 119+ vision models, filterable by organization, task, modality, and access type.
- [Playground](https://playground.roboflow.com/object-detection): Upload an image, enter a prompt, and compare outputs from multiple models side by side.
- [Arena](https://playground.roboflow.com/arena): Blind model battles. Vote on results without knowing which models you are comparing until after voting.
- [Rankings](https://playground.roboflow.com/ranking): ELO-based leaderboards showing which models perform best for each task type, updated in real time from Arena votes.
- [Vision Evals](https://playground.roboflow.com/evals): Ground-truth benchmark scores for vision-language models across object detection, counting, identification, OCR, data extraction, and reasoning.
- [Compare Models](https://playground.roboflow.com/models/compare): Select 2 to 4 models to compare specifications, supported tasks, and capabilities side by side.

## Playground

Task-specific testing at /{task-type}. Upload an image, enter a prompt, and compare outputs from multiple models side by side.

- [Object Detection](https://playground.roboflow.com/object-detection): Detect and localize objects in an image.
- [OCR](https://playground.roboflow.com/ocr): Extract text from an image.
- [Captioning](https://playground.roboflow.com/captioning): Generate a descriptive caption for an image.
- [PCS](https://playground.roboflow.com/pcs): Promptable Concept Segmentation.

## Arena

Blind comparisons at /arena or /arena/{task-type}. Upload an image for a blind model battle and vote on results. Votes directly update the ranking leaderboards.

- [Arena Home](https://playground.roboflow.com/arena): Start a blind model battle.
- [Arena: OCR](https://playground.roboflow.com/arena/ocr): Blind OCR model comparison.

## Rankings

ELO-based leaderboards at /ranking or /ranking/{task-type}. Each ranking page includes a "What is {task}?" section and an FAQ.

- [Rankings Home](https://playground.roboflow.com/ranking): Overall leaderboards across task types.
- [Rankings: Object Detection](https://playground.roboflow.com/ranking/object-detection): Object detection leaderboard.

## Vision Evals

Ground-truth benchmark scoring at /evals. Current vision-language models run the same real-world samples across six tasks, scored against ground truth: object detection (mAP), counting, identification, data extraction, and reasoning (accuracy), and OCR (mean similarity). Token usage, cost, and speed are measured per sample. Use Vision Evals alongside Arena Rankings: Evals show which model gets the right answer, Rankings show which model people prefer. The previous version of Vision Evals (deterministic pass/fail scoring of 50+ models, including deprecated ones that can no longer be re-run) is preserved at /evals/legacy.

- [Vision Evals](https://playground.roboflow.com/evals): The full Vision Evals leaderboard across all six tasks.
- [Object Detection Benchmark](https://playground.roboflow.com/evals/object-detection): Zero-shot detection mAP rankings.
- [Counting Benchmark](https://playground.roboflow.com/evals/object-counting): Object counting accuracy rankings.
- [Identification Benchmark](https://playground.roboflow.com/evals/visual-identification): Visual identification rankings.
- [OCR Benchmark](https://playground.roboflow.com/evals/ocr): Text transcription accuracy rankings.
- [Data Extraction Benchmark](https://playground.roboflow.com/evals/data-extraction): Targeted field extraction rankings.
- [Reasoning Benchmark](https://playground.roboflow.com/evals/visual-reasoning): Visual reasoning rankings.
- [Vision Evals (legacy)](https://playground.roboflow.com/evals/legacy): The previous pass/fail benchmark, preserved for reference.

## Deploying Models

Roboflow-provided models can be deployed as hosted REST APIs through Roboflow Workflows. Eligible model pages include a Deploy section with the endpoint contract, request schema, code examples in Python, JavaScript, and cURL, and a ready-to-paste integration prompt for AI coding agents. The general contract:

- Endpoint: POST https://serverless.roboflow.com/{workspace}/workflows/{workflow-id}
- Auth: send a Roboflow API key as api_key in the JSON request body. Get a key at https://app.roboflow.com/settings/api.
- Request body: { "api_key": "...", "inputs": { "image": { "type": "url" | "base64", "value": "..." } } }. Detection and classification workflows also take a classes string array in inputs; visual question answering workflows take a prompt string; workflows that route to an external VLM provider additionally take that provider's key as model_api_key.
- Forking: the "Deploy with an API" button on a model page forks the model's workflow into your own Roboflow workspace and fills the code examples with your real workflow ID and API key.
- AI coding agents (Claude Code, Cursor, and others) can connect to the Roboflow MCP server at https://mcp.roboflow.com/mcp to read workflow schemas and test-run workflows while integrating a model. Optional agent skills: https://github.com/roboflow/computer-vision-skills.

- [RF-DETR Deploy section](https://playground.roboflow.com/models/roboflow/rf-detr): Example model page with an on-page Deploy section, including per-size workflow IDs and the coding-agent prompt.
- [Roboflow Workflows documentation](https://docs.roboflow.com/workflows/what-is-workflows): Full Workflows reference.

## Model Comparisons

Direct model comparisons at /models/compare. Deep-link to a specific comparison at /models/compare/{model-a}-vs-{model-b}.

- [Compare Models](https://playground.roboflow.com/models/compare): Pick 2 to 4 models to compare.
- [Claude 4 Sonnet vs GPT-4o](https://playground.roboflow.com/models/compare/claude-4-sonnet-vs-gpt-4o): Example head-to-head comparison.
- [Claude 4 Opus vs Gemini 2.5 Pro](https://playground.roboflow.com/models/compare/claude-4-opus-vs-gemini-2.5-pro): Example head-to-head comparison.

## Organizations

Organization pages list every model from a single provider at /models/{organization-slug}.

- [Google](https://playground.roboflow.com/models/google): Models from Google.
- [Anthropic](https://playground.roboflow.com/models/anthropic): Models from Anthropic.
- [OpenAI](https://playground.roboflow.com/models/openai): Models from OpenAI.

## Supported Tasks

Each task slug below is valid for a /models/task/{slug} landing page, which lists the top models for that task.

- [Captioning](https://playground.roboflow.com/models/task/captioning): Models that generate descriptive captions for images.
- [Classification](https://playground.roboflow.com/models/task/classification): Models that classify images into categories.
- [Instance Segmentation (COCO)](https://playground.roboflow.com/models/task/instance-segmentation-coco): Compare pretrained instance segmentation models on the 80 COCO classes. No class list needed.
- [Object Detection](https://playground.roboflow.com/models/task/object-detection): Models that detect and localize objects in images.
- [Object Detection (COCO)](https://playground.roboflow.com/models/task/object-detection-coco): Compare pretrained object detection models on the 80 COCO classes. No class list needed.
- [OCR](https://playground.roboflow.com/models/task/ocr): Models that extract text from images.
- [Open Prompt](https://playground.roboflow.com/models/task/open-prompt): Models that interpret free-form prompts on images.
- [Promptable Concept Segmentation](https://playground.roboflow.com/models/task/pcs)

## Supported Features

Each feature slug below is valid for a /models/feature/{slug} landing page, which lists the top models sharing that feature.

- [Foundation Vision](https://playground.roboflow.com/models/feature/foundation-vision)
- [LLMs with Vision Capabilities](https://playground.roboflow.com/models/feature/llms-with-vision-capabilities)
- [Multimodal Vision](https://playground.roboflow.com/models/feature/multimodal-vision)
- [Real-Time Vision](https://playground.roboflow.com/models/feature/real-time-vision)
- [Zero-shot Detection](https://playground.roboflow.com/models/feature/zero-shot-detection)

## Supported Models

Each model links to its detail page at /models/{organization-slug}/{model-slug}.

- [ByteTrack](https://playground.roboflow.com/models/bytedance/bytetrack)
- [Claude Fable 5](https://playground.roboflow.com/models/anthropic/claude-fable-5)
- [Claude Haiku 4.5](https://playground.roboflow.com/models/anthropic/claude-4-5-haiku)
- [Claude Opus 4.1](https://playground.roboflow.com/models/anthropic/claude-4-1-opus)
- [Claude Opus 4.5](https://playground.roboflow.com/models/anthropic/claude-4-5-opus)
- [Claude Opus 4.6](https://playground.roboflow.com/models/anthropic/claude-opus-4-6)
- [Claude Opus 4.7](https://playground.roboflow.com/models/anthropic/claude-opus-4-7)
- [Claude Opus 4.8](https://playground.roboflow.com/models/anthropic/claude-opus-4-8)
- [Claude Sonnet 4.5](https://playground.roboflow.com/models/anthropic/claude-4-5-sonnet)
- [Claude Sonnet 4.6](https://playground.roboflow.com/models/anthropic/claude-sonnet-4-6)
- [Claude Sonnet 5](https://playground.roboflow.com/models/anthropic/claude-sonnet-5)
- [CLIP](https://playground.roboflow.com/models/openai/openai-clip)
- [Co-DETR](https://playground.roboflow.com/models/openmmlab/co-detr)
- [D-FINE](https://playground.roboflow.com/models/ustc/d-fine)
- [DEIM](https://playground.roboflow.com/models/intellindust-ai-lab/deim)
- [Depth Anything V2](https://playground.roboflow.com/models/bytedance/depth-anything-v2)
- [Detectron2](https://playground.roboflow.com/models/meta/detectron2)
- [DETR](https://playground.roboflow.com/models/meta/detr)
- [DINOv2](https://playground.roboflow.com/models/meta/dinov2)
- [docTR](https://playground.roboflow.com/models/mindee/doctr)
- [EfficientDet](https://playground.roboflow.com/models/google/efficientdet)
- [Faster R-CNN](https://playground.roboflow.com/models/microsoft/faster-r-cnn)
- [Florence-2](https://playground.roboflow.com/models/microsoft/florence-2)
- [Gemini 2.5 Flash](https://playground.roboflow.com/models/google/gemini-2-5-flash)
- [Gemini 2.5 Flash-Lite](https://playground.roboflow.com/models/google/gemini-2-5-flash-lite)
- [Gemini 2.5 Pro](https://playground.roboflow.com/models/google/gemini-2-5-pro)
- [Gemini 3 Flash](https://playground.roboflow.com/models/google/gemini-3-flash)
- [Gemini 3.1 Flash-Lite](https://playground.roboflow.com/models/google/gemini-3-1-flash-lite)
- [Gemini 3.1 Pro](https://playground.roboflow.com/models/google/gemini-3-1-pro)
- [Gemini 3.5 Flash](https://playground.roboflow.com/models/google/gemini-3-5-flash)
- [Gemma 3 12B](https://playground.roboflow.com/models/google/gemma-3-12b)
- [Gemma 3 27B](https://playground.roboflow.com/models/google/gemma-3-27b)
- [Gemma 3 4B](https://playground.roboflow.com/models/google/gemma-3-4b)
- [Gemma 4 12B](https://playground.roboflow.com/models/google/gemma-4-12b)
- [Gemma 4 26B A4B](https://playground.roboflow.com/models/google/gemma-4-26b-a4b)
- [Gemma 4 31B](https://playground.roboflow.com/models/google/gemma-4-31b)
- [GLM-OCR](https://playground.roboflow.com/models/z-ai/glm-ocr)
- [Google Vision OCR](https://playground.roboflow.com/models/google/google-vision-ocr)
- [GPT-5](https://playground.roboflow.com/models/openai/gpt-5)
- [GPT-5 Mini](https://playground.roboflow.com/models/openai/gpt-5-mini)
- [GPT-5 Nano](https://playground.roboflow.com/models/openai/gpt-5-nano)
- [GPT-5.1](https://playground.roboflow.com/models/openai/gpt-5-1)
- [GPT-5.2](https://playground.roboflow.com/models/openai/gpt-5-2)
- [GPT-5.4](https://playground.roboflow.com/models/openai/gpt-5-4)
- [GPT-5.4 Mini](https://playground.roboflow.com/models/openai/gpt-5-4-mini)
- [GPT-5.4 Nano](https://playground.roboflow.com/models/openai/gpt-5-4-nano)
- [GPT-5.5](https://playground.roboflow.com/models/openai/gpt-5-5)
- [GPT-5.6 Luna](https://playground.roboflow.com/models/openai/gpt-5-6-luna)
- [GPT-5.6 Sol](https://playground.roboflow.com/models/openai/gpt-5-6-sol)
- [GPT-5.6 Terra](https://playground.roboflow.com/models/openai/gpt-5-6-terra)
- [Grok 4](https://playground.roboflow.com/models/xai/grok-4)
- [Grounded SAM](https://playground.roboflow.com/models/idea-research/grounded-sam)
- [Grounding DINO](https://playground.roboflow.com/models/idea-research/grounding-dino)
- [Kimi K2.5](https://playground.roboflow.com/models/moonshot-ai/kimi-k2-5)
- [Llama 3.2 Vision 11b](https://playground.roboflow.com/models/meta/llama-3-2-vision-11b)
- [Llama 4 Maverick](https://playground.roboflow.com/models/meta/llama-4-maverick)
- [Llama 4 Scout](https://playground.roboflow.com/models/meta/llama-4-scout)
- [LLaVA-1.5](https://playground.roboflow.com/models/microsoft/llava-1-5)
- [Mask R-CNN](https://playground.roboflow.com/models/meta/mask-rcnn)
- [MediaPipe](https://playground.roboflow.com/models/google/mediapipe)
- [Mistral Medium 3.1](https://playground.roboflow.com/models/mistral/mistral-medium-3-1)
- [Mistral Small 3.1 24B](https://playground.roboflow.com/models/mistral/mistral-small-3-1-24b)
- [MobileNet SSD v2](https://playground.roboflow.com/models/google/mobilenet-ssd-v2)
- [MobileNetV2](https://playground.roboflow.com/models/google/mobilenetv2)
- [Moondream 2](https://playground.roboflow.com/models/moondream-ai/moondream-2)
- [OWL-ViT](https://playground.roboflow.com/models/google/owl-vit)
- [PaliGemma](https://playground.roboflow.com/models/google/paligemma)
- [PaliGemma 2](https://playground.roboflow.com/models/google/paligemma-2)
- [Pixtral 12B](https://playground.roboflow.com/models/mistral/pixtral-12b)
- [Qwen VL Max](https://playground.roboflow.com/models/qwen/qwen-vl-max)
- [Qwen-VL](https://playground.roboflow.com/models/qwen/qwen-vl)
- [Qwen2.5 VL 7B Instruct](https://playground.roboflow.com/models/qwen/qwen2-5-vl-7b-instruct)
- [Qwen3 VL 235B A22B Instruct](https://playground.roboflow.com/models/qwen/qwen3-vl-235b-a22b-instruct)
- [Qwen3 VL 30B A3B Instruct](https://playground.roboflow.com/models/qwen/qwen3-vl-30b-a3b-instruct)
- [Qwen3 VL 8B Instruct](https://playground.roboflow.com/models/qwen/qwen3-vl-8b-instruct)
- [Qwen3.5 122B A10B](https://playground.roboflow.com/models/qwen/qwen3-5-122b-a10b)
- [Qwen3.5 27B](https://playground.roboflow.com/models/qwen/qwen3-5-27b)
- [Qwen3.5 35B A3B](https://playground.roboflow.com/models/qwen/qwen3-5-35b-a3b)
- [Qwen3.5 397B A17B](https://playground.roboflow.com/models/qwen/qwen3-5-397b-a17b)
- [Qwen3.5 9b](https://playground.roboflow.com/models/qwen/qwen3-5-9b)
- [Qwen3.6 27B](https://playground.roboflow.com/models/qwen/qwen3-6-27b)
- [Qwen3.6 35B A3B](https://playground.roboflow.com/models/qwen/qwen3-6-35b-a3b)
- [Qwen3.6 Flash](https://playground.roboflow.com/models/qwen/qwen3-6-flash)
- [Qwen3.6 Plus](https://playground.roboflow.com/models/qwen/qwen3-6-plus)
- [ResNet-32](https://playground.roboflow.com/models/meta/resnet-32)
- [ResNet-34](https://playground.roboflow.com/models/meta/resnet-34)
- [ResNet-50](https://playground.roboflow.com/models/microsoft/resnet-50)
- [RF-DETR](https://playground.roboflow.com/models/roboflow/rf-detr)
- [RF-DETR Segmentation](https://playground.roboflow.com/models/roboflow/rf-detr-segmentation)
- [RT-DETR](https://playground.roboflow.com/models/baidu/rt-detr)
- [RTMDet](https://playground.roboflow.com/models/openmmlab/rtmdet)
- [SAM 3](https://playground.roboflow.com/models/meta/sam-3)
- [SAM 3D Objects](https://playground.roboflow.com/models/meta/sam-3d-objects)
- [SAM-CLIP](https://playground.roboflow.com/models/apple/sam-clip)
- [Segment Anything Model (SAM)](https://playground.roboflow.com/models/meta/sam)
- [Segment Anything Model 2 (SAM 2)](https://playground.roboflow.com/models/meta/sam-2)
- [SigLIP](https://playground.roboflow.com/models/google/siglip)
- [SmolVLM2](https://playground.roboflow.com/models/hugging-face/smolvlm2)
- [Surya](https://playground.roboflow.com/models/mindee/surya)
- [TrOCR](https://playground.roboflow.com/models/microsoft/trocr)
- [Vision Transformer (ViT)](https://playground.roboflow.com/models/google/vision-transformer-vit)
- [YOLO World](https://playground.roboflow.com/models/tencent-ai-lab/yolo-world)
- [YOLO-NAS](https://playground.roboflow.com/models/deci-ai/yolo-nas)
- [YOLO11](https://playground.roboflow.com/models/ultralytics/yolo11)
- [YOLO26](https://playground.roboflow.com/models/ultralytics/yolo26)
- [YOLOE](https://playground.roboflow.com/models/thu-mig/yoloe)
- [YOLOS](https://playground.roboflow.com/models/hugging-face/yolos)
- [YOLOv10](https://playground.roboflow.com/models/thu-mig/yolov10)
- [YOLOv12](https://playground.roboflow.com/models/thu-mig/yolov12)
- [YOLOv4](https://playground.roboflow.com/models/academia-sinica/yolov4-darknet)
- [YOLOv4-tiny](https://playground.roboflow.com/models/academia-sinica/yolov4-tiny)
- [YOLOv5](https://playground.roboflow.com/models/ultralytics/yolov5)
- [YOLOv7](https://playground.roboflow.com/models/academia-sinica/yolov7)
- [YOLOv8](https://playground.roboflow.com/models/ultralytics/yolov8)
- [YOLOv8 Classification](https://playground.roboflow.com/models/ultralytics/yolov8-classification)
- [YOLOv8 Instance Segmentation](https://playground.roboflow.com/models/ultralytics/yolov8-instance-segmentation)
- [YOLOv8 Pose Estimation](https://playground.roboflow.com/models/ultralytics/yolov8-pose-estimation)
- [YOLOv9](https://playground.roboflow.com/models/academia-sinica/yolov9)
- [YOLOX](https://playground.roboflow.com/models/megvii/yolox)

## Keywords

- compare vision models side by side
- vision model arena
- AI model leaderboard
- blind model comparison
- ELO ranking system
- vision language models (VLM)
- test vision models free no login
- Claude vs GPT vs Gemini vision
- AI vision model directory
- computer vision models
- promptable concept segmentation (PCS)
- PCS leaderboard
- OCR AI
- OCR leaderboard
- object detection leaderboard
- captioning leaderboard
- classification leaderboard
- VQA leaderboard
- vision model accuracy eval
- vision eval leaderboard
- deterministic vision model testing
- vision model pass fail eval
- OCR accuracy eval
- object detection accuracy eval
- document understanding eval
- deploy vision model API
- Roboflow Workflows API
- Roboflow inference API

## Audience

- AI researchers
- ML engineers
- developers building vision apps
- educators
- AI enthusiasts and learners

---

_Maintainer: Roboflow. Contact: chris@roboflow.com. Last updated: 2026-07-14._